What changed between 5 Sept 2026 and 12 Sept 2026
Two dates, one scope: new and retired models, price and context changes, new benchmark leaders, new papers, provider and hardware changes. Built from the change log keyed on when things occurred — nothing is inferred, and the URL is the report.
Scope: the whole atlas
9,980 events in the window · 1,852 claims superseded · price rows opened 640 / closed 20 · entities at 5 Sept 2026: 0 → at 12 Sept 2026: 5,944. The entity lists are capped at 200 by the API — per-type counts below are “of the first 200”. Events are keyed on occurred_at (effective date when known) and exclude back-filled history unless include_backfill=1; new_entities uses first_seen_at. new_benchmark_leaders compares the primary-group leader computed from results observed by each date.
New models
New models 73
| Model | Organization | Params | Context | Openness | Released | First seen |
|---|---|---|---|---|---|---|
| Phi-4-mini-instruct (3.8B), SigLIP2-so400M | — | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Omni | Qwen Team | 7B | — | Open weights | 27 Mar 2025 | 12 Sept 2026 |
| Qwen2.5-Omni-7B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| V4-Flash | DeepSeek | — | — | — | — | 12 Sept 2026 |
| qwen-mt-turbo | Qwen Team | — | — | Proprietary | 24 Jul 2025 | 12 Sept 2026 |
| Document AI | — | — | — | — | — | 12 Sept 2026 |
| Mistral OCR 3 | — | — | — | — | — | 12 Sept 2026 |
| Mistral OCR 4 | Mistral AI | — | — | Proprietary | 23 Jun 2026 | 12 Sept 2026 |
| deepseek-chat, deepseek-reasoner | — | — | — | — | — | 12 Sept 2026 |
| DeepSeek-V4 | DeepSeek | 1.6T | 1M | Open weights | — | 12 Sept 2026 |
| Qwen-VL | — | — | — | — | — | 12 Sept 2026 |
| Qwen2-VL-7B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2-VL-2B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen-Coder | — | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Coder-Base | — | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Coder-14B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Coder-3B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Coder-0.5B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Vibe Enterprise | — | — | — | — | — | 12 Sept 2026 |
| Vibe Team | — | — | — | — | — | 12 Sept 2026 |
| Vibe Pro | — | — | — | — | — | 12 Sept 2026 |
| Vibe | — | — | — | — | — | 12 Sept 2026 |
| Qwen-TTS | Qwen Team | — | — | Proprietary | 27 Jun 2025 | 12 Sept 2026 |
| qwen-tts-2025-05-22 | Qwen Team | — | — | — | — | 12 Sept 2026 |
| qwen-tts-latest | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen3 Embedding Series | Qwen Team | — | 32.8K | Open weights | 5 Jun 2025 | 12 Sept 2026 |
| Qwen3-Reranker-8B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen3-Reranker-4B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen3-Reranker-0.6B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen3-Embedding-8B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen3-Embedding-4B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| DeepSeek-R1-0528 | DeepSeek | — | — | Open weights | — | 12 Sept 2026 |
| DeepSeek-V3.1-Think | DeepSeek | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-VL series | — | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-VL-32B-Instruct | Qwen Team | 32B | — | Open source | 24 Mar 2025 | 12 Sept 2026 |
| Vision-language model specialized for grounding tasks such as pointing, counting, and object localization | — | — | — | — | — | 12 Sept 2026 |
| Robostral Navigate | Mistral AI | 8B | — | Proprietary | — | 12 Sept 2026 |
| Qwen2-VL | Qwen Team | — | — | Open weights | 29 Aug 2024 | 12 Sept 2026 |
| Qwen2.5-VL-3B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Opus 4.8 | — | — | — | — | — | 12 Sept 2026 |
| Qwen2.5 VL | Qwen Team | — | — | Open weights | 26 Jan 2025 | 12 Sept 2026 |
| Qwen VLo | Qwen Team | — | — | unknown | 26 Jun 2025 | 12 Sept 2026 |
| Qwen2-Math | — | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Math-RM-72B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Math-7B-Instruct | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen-Audio | — | — | — | — | — | 12 Sept 2026 |
| Qwen | — | — | — | — | — | 12 Sept 2026 |
| Qwen2-Audio | Qwen Team | 7B | — | Open weights | 9 Aug 2024 | 12 Sept 2026 |
| Qwen2-Audio-7B-Instruct | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2-Audio-7B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| QwQ-Max-Preview | Qwen Team | — | — | Open source | 25 Feb 2025 | 12 Sept 2026 |
| Qwen-Image-Edit | Qwen Team | 20B | — | unknown | 19 Aug 2025 | 12 Sept 2026 |
| CodeQwen1.5 | — | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Coder-Instruct | Qwen Team | — | — | — | — | 12 Sept 2026 |
| V3.1-Terminus | — | — | — | — | — | 12 Sept 2026 |
| V3.2-Exp | — | — | — | — | — | 12 Sept 2026 |
| QVQ-72B-Preview | Qwen Team | 72B | — | Open weights | 25 Dec 2024 | 12 Sept 2026 |
| QVQ-Max | Qwen Team | — | — | Proprietary | 28 Mar 2025 | 12 Sept 2026 |
| Qwen2.5-1M | Qwen Team | — | 1M | Open source | 27 Jan 2025 | 12 Sept 2026 |
| Qwen2.5-14B-Instruct-1M | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-7B-Instruct-1M | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2 | — | — | — | — | — | 12 Sept 2026 |
| Qwen2-VL-72B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Math-1.5B-Instruct | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Math-1.5B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Math-72B-Instruct | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Math-72B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Coder-1.5B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-14B-Instruct | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-14B | Qwen Team | — | — | — | — | 12 Sept 2026 |
| Qwen2.5-Math | Qwen Team | — | 4.1K | Open source | 19 Sept 2024 | 12 Sept 2026 |
| Qwen2.5-Coder | Qwen Team | 32.5B | 128K | Open source | 12 Nov 2024 | 12 Sept 2026 |
| Qwen2.5 | Qwen Team | 72.7B | 131.1K | Open weights | 19 Sept 2024 | 12 Sept 2026 |
Retired models
Retired models 0
No model was retired between these dates
Price changes
Price changes 20
| Date | Model | Provider | Input / 1M | Output / 1M | Source |
|---|---|---|---|---|---|
| DeepSeek V4 Flash (0731) | DeepSeek API | $0.065→$0.04−38.5% | $0.18→$0.08−55.6% | openrouter.ai | |
| DeepSeek V4 Pro 0423 | DeepSeek API | $0.829→$0.819−1.2% | $1.66→$1.64−1.2% | openrouter.ai | |
| DeepSeek V4 Flash 0423 | DeepSeek API | $0.067→$0.067−0.4% | $0.134→$0.134−0.4% | openrouter.ai | |
| DeepSeek V4 Flash 0423 | DeepSeek API | $0.067→$0.067+0.4% | $0.134→$0.134+0.4% | openrouter.ai | |
| DeepSeek V4 Pro 0423 | DeepSeek API | $0.85→$0.829−2.4% | $1.7→$1.66−2.4% | openrouter.ai | |
| DeepSeek V4 Flash 0423 | DeepSeek API | $0.067→$0.067−0.8% | $0.135→$0.134−0.8% | openrouter.ai | |
| Qwen3.8 27B | OpenRouter | $0.42→$0.214−49.0% | $3→$2.55−15.0% | openrouter.ai | |
| DeepSeek-V4-Pro-0813 | DeepSeek API | $0.579→$0.578−0.2% | $1.74→$1.73−0.2% | openrouter.ai | |
| Kimi K3 | Moonshot AI Platform | $1.54→$2.3+50.0% | $7.7→$11.55+50.0% | openrouter.ai | |
| DeepSeek V4 Pro 0423 | DeepSeek API | $0.84→$0.85+1.2% | $1.68→$1.7+1.2% | openrouter.ai | |
| DeepSeek V4 Flash 0423 | DeepSeek API | $0.068→$0.067−0.4% | $0.135→$0.135−0.4% | openrouter.ai | |
| MiniMax M2.5 | MiniMax API | $0.27→$0.30+11.1% | $1.08→$1.2+11.1% | openrouter.ai | |
| Qwen3 235B A22B Instruct 2507 | OpenRouter | $0.22→$0.087−60.2% | $0.88→$0.35−60.2% | openrouter.ai | |
| Llama 3.1 70B Instruct | OpenRouter | $0.40→$0.72+80.0% | $0.40→$0.72+80.0% | openrouter.ai | |
| Kimi K3 | Moonshot AI Platform | $1.8→$1.54−14.5% | $9.01→$7.7−14.5% | openrouter.ai | |
| Hy3 | OpenRouter | $0.083→$0.132+60.0% | $0.33→$0.528+60.0% | openrouter.ai | |
| DeepSeek V4 Pro 0423 | DeepSeek API | $0.948→$0.84−11.4% | $1.9→$1.68−11.4% | openrouter.ai | |
| DeepSeek V4 Flash 0423 | DeepSeek API | $0.085→$0.068−20.0% | $0.169→$0.135−20.0% | openrouter.ai | |
| DeepSeek V4 Pro 0423 | DeepSeek API | $0.86→$0.948+10.2% | $1.72→$1.9+10.2% | openrouter.ai | |
| DeepSeek V4 Flash 0423 | DeepSeek API | $0.085→$0.085−0.3% | $0.17→$0.169−0.3% | openrouter.ai |
Context changes
Context changes 182
| Date | Model | Organization | Context window | Source | History |
|---|---|---|---|---|---|
| Qwen3 235B A22B | Qwen | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| Gemma 3 27B | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | ||
| Ling 3.0 Flash VL | inclusionAI | 262.1K tokens→131.1K tokens−50.0% | openrouter.ai | claims → | |
| Muse Spark 1.3 | Meta AI | 1M tokens→1.05M tokens+4.9% | openrouter.ai | claims → | |
| Qwen3.8 Flash | Qwen | 256K tokens→1M tokens+290.6% | openrouter.ai | claims → | |
| GLM 5.3 Flash | Z.ai (Zhipu AI) | 1M tokens→1.31M tokens+31.1% | openrouter.ai | claims → | |
| GLM 5.3 | Z.ai (Zhipu AI) | 1M tokens→1.31M tokens+31.1% | openrouter.ai | claims → | |
| Qwen3.8 27B | Qwen | 256K tokens→1M tokens+290.6% | openrouter.ai | claims → | |
| Qwen3.8 2.4T A95B | Qwen | 983.6K tokens→1.05M tokens+6.6% | openrouter.ai | claims → | |
| Nemotron 3.5 Lightning | NVIDIA | 1M tokens→262.1K tokens−73.8% | openrouter.ai | claims → | |
| Solar Pro 4 | Upstage | 512K tokens→524.3K tokens+2.4% | openrouter.ai | claims → | |
| Inkling Small | Thinking Machines | 1M tokens→1.05M tokens+4.9% | openrouter.ai | claims → | |
| LongCat 2.0 | Meituan | 1M tokens→1.05M tokens+4.9% | openrouter.ai | claims → | |
| Inkling | Thinking Machines | 1M tokens→1.05M tokens+4.9% | openrouter.ai | claims → | |
| Hy3 | Tencent | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Kimi K2.7 Code | Moonshot AI | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| MiniMax M3 | MiniMax | 1M tokens→1.05M tokens+4.9% | openrouter.ai | claims → | |
| Qwen3.6 Max Preview | Qwen | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| gpt-5.5-pro | OpenAI | 922K tokens→1.05M tokens+13.9% | openrouter.ai | claims → | |
| gpt-5.5 | OpenAI | 922K tokens→1.05M tokens+13.9% | openrouter.ai | claims → | |
| Hy3 preview | Tencent | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| MiMo-V2.5-Pro | Xiaomi | 1M tokens→1.05M tokens+5.0% | openrouter.ai | claims → | |
| MiMo-V2.5 | Xiaomi | 1M tokens→1.05M tokens+5.0% | openrouter.ai | claims → | |
| Kimi K2.6 | Moonshot AI | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| GLM 5.1 | Z.ai (Zhipu AI) | 200K tokens→204.8K tokens+2.4% | openrouter.ai | claims → | |
| Gemma 4 26B A4B | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | ||
| Gemma 4 31B | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | ||
| GLM 5V Turbo | Z.ai (Zhipu AI) | 200K tokens→202.8K tokens+1.4% | openrouter.ai | claims → | |
| Trinity Large Thinking | Arcee AI | 512K tokens→262.1K tokens−48.8% | openrouter.ai | claims → | |
| KAT-Coder-Pro V2 | Kwaipilot | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| GLM 5 Turbo | Z.ai (Zhipu AI) | 200K tokens→202.8K tokens+1.4% | openrouter.ai | claims → | |
| Nemotron 3 Super | NVIDIA | 1M tokens→262.1K tokens−73.8% | openrouter.ai | claims → | |
| GLM 5 | Z.ai (Zhipu AI) | 200K tokens→204.8K tokens+2.4% | openrouter.ai | claims → | |
| Qwen3 Max Thinking | Qwen | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Qwen3 Coder Next | Qwen | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Step 3.5 Flash | StepFun | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Kimi K2.5 | Moonshot AI | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Solar Pro 3 | Upstage | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| GLM 4.7 | Z.ai (Zhipu AI) | 200K tokens→204.8K tokens+2.4% | openrouter.ai | claims → | |
| Nemotron 3 Nano 30B A3B | NVIDIA | 1M tokens→262.1K tokens−73.8% | openrouter.ai | claims → | |
| Devstral 2 | Mistral AI | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| GLM 4.6V | Z.ai (Zhipu AI) | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| DeepSeek V3.2 | DeepSeek | 128K tokens→163.8K tokens+28.0% | openrouter.ai | claims → | |
| gpt-5.1 | OpenAI | 272K tokens→400K tokens+47.1% | openrouter.ai | claims → | |
| Kimi K2 Thinking | Moonshot AI | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Qwen3 VL 32B Instruct | Qwen | 256K tokens→131.1K tokens−48.8% | openrouter.ai | claims → | |
| Granite 4.0 Micro | IBM | 128K tokens→131K tokens+2.3% | openrouter.ai | claims → | |
| Qwen3 VL 8B Instruct | Qwen | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Qwen3 VL 30B A3B Instruct | Qwen | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| GLM 4.6 | Z.ai (Zhipu AI) | 200K tokens→204.8K tokens+2.4% | openrouter.ai | claims → | |
| DeepSeek V3.2 Exp | DeepSeek | 128K tokens→163.8K tokens+28.0% | openrouter.ai | claims → | |
| DeepSeek V3.1 Terminus | DeepSeek | 128K tokens→163.8K tokens+28.0% | openrouter.ai | claims → | |
| Kimi K2 0905 | Moonshot AI | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Hermes 4 405B | Nous Research | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| DeepSeek V3.1 | DeepSeek | 128K tokens→163.8K tokens+28.0% | openrouter.ai | claims → | |
| Mistral Medium 3.1 | Mistral AI | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| GLM 4.5V | Z.ai (Zhipu AI) | 64K tokens→65.5K tokens+2.4% | openrouter.ai | claims → | |
| GLM 4.5 | Z.ai (Zhipu AI) | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| GLM 4.5 Air | Z.ai (Zhipu AI) | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| Qwen3 235B A22B Instruct 2507 | Qwen | 256K tokens→262.1K tokens+2.4% | openrouter.ai | claims → | |
| Kimi K2 0711 | Moonshot AI | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| Mistral Medium 3 | Mistral AI | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| Qwen3 14B | Qwen | 32.8K tokens→131.1K tokens+300.0% | openrouter.ai | claims → | |
| Qwen3 32B | Qwen | 32.8K tokens→131.1K tokens+300.0% | openrouter.ai | claims → | |
| gpt-4.1 | OpenAI | 1M tokens→1.05M tokens+4.8% | openrouter.ai | claims → | |
| gpt-4.1-mini | OpenAI | 1M tokens→1.05M tokens+4.8% | openrouter.ai | claims → | |
| gpt-4.1-nano | OpenAI | 1M tokens→1.05M tokens+4.8% | openrouter.ai | claims → | |
| Llama 4 Maverick | Meta AI | 1M tokens→1.05M tokens+4.9% | openrouter.ai | claims → | |
| Llama 4 Scout | Meta AI | 10M tokens→1.31M tokens−86.9% | openrouter.ai | claims → | |
| DeepSeek V3 0324 | DeepSeek | 128K tokens→163.8K tokens+28.0% | openrouter.ai | claims → | |
| Gemma 3 4B | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | ||
| Gemma 3 12B | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | ||
| Reka Flash 3 | rekaai | 128K tokens→65.5K tokens−48.8% | openrouter.ai | claims → | |
| Sonar Reasoning Pro | Perplexity AI | 127K tokens→128K tokens+0.8% | openrouter.ai | claims → | |
| Mistral Saba | Mistral AI | 32K tokens→32.8K tokens+2.4% | openrouter.ai | claims → | |
| Mistral Small 3 | Mistral AI | 32K tokens→32.8K tokens+2.4% | openrouter.ai | claims → | |
| Sonar | Perplexity AI | 127K tokens→127.1K tokens+0.1% | openrouter.ai | claims → | |
| R1 Distill Llama 70B | DeepSeek | 128K tokens→8.19K tokens−93.6% | openrouter.ai | claims → | |
| Phi 4 | Microsoft | 16K tokens→16.4K tokens+2.4% | openrouter.ai | claims → | |
| DeepSeek V3 | DeepSeek | 128K tokens→163.8K tokens+28.0% | openrouter.ai | claims → | |
| Mistral Large 2.0 | Mistral AI | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| Qwen2.5 Coder 32B Instruct | Qwen | 131.1K tokens→32.8K tokens−75.0% | openrouter.ai | claims → | |
| Hermes 3 70B Instruct | Nous Research | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| Mistral Large | Mistral AI | 32.8K tokens→128K tokens+290.6% | openrouter.ai | claims → | |
| gpt-3.5-turbo | OpenAI | 4.1K tokens→4.1K tokens−0.0% | openrouter.ai | claims → | |
| GPT-3.5 Turbo | OpenAI | 4.1K tokens→16.4K tokens+300.0% | openrouter.ai | claims → | |
| gpt-4 | OpenAI | 8.19K tokens→8.19K tokens−0.0% | openrouter.ai | claims → | |
| GPT-3.5 Turbo | OpenAI | 16.4K tokens→4.1K tokens−75.0% | artificialanalysis.ai | claims → | |
| Mistral Large 2.0 | Mistral AI | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| GLM 5.3 Flash | Z.ai (Zhipu AI) | 1.31M tokens→1M tokens−23.7% | artificialanalysis.ai | claims → | |
| GLM 5 Turbo | Z.ai (Zhipu AI) | 202.8K tokens→200K tokens−1.4% | artificialanalysis.ai | claims → | |
| Kimi K2.6 | Moonshot AI | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Inkling | Thinking Machines | 1.05M tokens→1M tokens−4.6% | artificialanalysis.ai | claims → | |
| Devstral 2 | Mistral AI | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Mistral Medium 3.1 | Mistral AI | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| Qwen2.5 Coder 32B Instruct | Qwen | 32.8K tokens→131.1K tokens+300.0% | artificialanalysis.ai | claims → | |
| MiMo-V2.5-Pro | Xiaomi | 1.05M tokens→1M tokens−4.8% | artificialanalysis.ai | claims → | |
| Qwen3 Coder Next | Qwen | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| KAT-Coder-Pro V2 | Kwaipilot | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| DeepSeek V3.1 | DeepSeek | 163.8K tokens→128K tokens−21.9% | artificialanalysis.ai | claims → | |
| Qwen3 235B A22B Instruct 2507 | Qwen | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Phi 4 | Microsoft | 16.4K tokens→16K tokens−2.3% | artificialanalysis.ai | claims → | |
| Mistral Small 3 | Mistral AI | 32.8K tokens→32K tokens−2.3% | artificialanalysis.ai | claims → | |
| gpt-5.5-pro | OpenAI | 1.05M tokens→922K tokens−12.2% | artificialanalysis.ai | claims → | |
| DeepSeek V3.1 Terminus | DeepSeek | 163.8K tokens→128K tokens−21.9% | artificialanalysis.ai | claims → | |
| Reka Flash 3 | rekaai | 65.5K tokens→128K tokens+95.3% | artificialanalysis.ai | claims → | |
| DeepSeek V3.2 | DeepSeek | 163.8K tokens→128K tokens−21.9% | artificialanalysis.ai | claims → | |
| GLM 4.6V | Z.ai (Zhipu AI) | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| GLM 5.3 | Z.ai (Zhipu AI) | 1.31M tokens→1M tokens−23.7% | artificialanalysis.ai | claims → | |
| Gemma 4 31B | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | ||
| Step 3.5 Flash | StepFun | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Kimi K2 Thinking | Moonshot AI | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Solar Pro 3 | Upstage | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| Trinity Large Thinking | Arcee AI | 262.1K tokens→512K tokens+95.3% | artificialanalysis.ai | claims → | |
| Hy3 | Tencent | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Qwen3.8 27B | Qwen | 1M tokens→256K tokens−74.4% | artificialanalysis.ai | claims → | |
| Qwen3 32B | Qwen | 131.1K tokens→32.8K tokens−75.0% | artificialanalysis.ai | claims → | |
| DeepSeek V3.2 Exp | DeepSeek | 163.8K tokens→128K tokens−21.9% | artificialanalysis.ai | claims → | |
| Llama 4 Scout | Meta AI | 1.31M tokens→10M tokens+662.9% | artificialanalysis.ai | claims → | |
| Kimi K2.5 | Moonshot AI | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| gpt-4.1-mini | OpenAI | 1.05M tokens→1M tokens−4.5% | artificialanalysis.ai | claims → | |
| Qwen3.8 Flash | Qwen | 1M tokens→256K tokens−74.4% | artificialanalysis.ai | claims → | |
| Muse Spark 1.3 | Meta AI | 1.05M tokens→1M tokens−4.6% | artificialanalysis.ai | claims → | |
| GLM 4.6 | Z.ai (Zhipu AI) | 204.8K tokens→200K tokens−2.3% | artificialanalysis.ai | claims → | |
| Llama 4 Maverick | Meta AI | 1.05M tokens→1M tokens−4.6% | artificialanalysis.ai | claims → | |
| Qwen3.8 2.4T A95B | Qwen | 1.05M tokens→983.6K tokens−6.2% | artificialanalysis.ai | claims → | |
| Kimi K2.7 Code | Moonshot AI | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Hy3 preview | Tencent | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| gpt-3.5-turbo | OpenAI | 4.1K tokens→4.1K tokens+0.0% | artificialanalysis.ai | claims → | |
| Hermes 4 405B | Nous Research | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| Qwen3 Max Thinking | Qwen | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Mistral Saba | Mistral AI | 32.8K tokens→32K tokens−2.3% | artificialanalysis.ai | claims → | |
| DeepSeek V3 0324 | DeepSeek | 163.8K tokens→128K tokens−21.9% | artificialanalysis.ai | claims → | |
| gpt-4.1-nano | OpenAI | 1.05M tokens→1M tokens−4.5% | artificialanalysis.ai | claims → | |
| Inkling Small | Thinking Machines | 1.05M tokens→1M tokens−4.6% | artificialanalysis.ai | claims → | |
| GLM 4.7 | Z.ai (Zhipu AI) | 204.8K tokens→200K tokens−2.3% | artificialanalysis.ai | claims → | |
| Gemma 4 26B A4B | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | ||
| Qwen3 VL 32B Instruct | Qwen | 131.1K tokens→256K tokens+95.3% | artificialanalysis.ai | claims → | |
| DeepSeek V3 | DeepSeek | 163.8K tokens→128K tokens−21.9% | artificialanalysis.ai | claims → | |
| gpt-4 | OpenAI | 8.19K tokens→8.19K tokens+0.0% | artificialanalysis.ai | claims → | |
| Kimi K2 0905 | Moonshot AI | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Nemotron 3 Super | NVIDIA | 262.1K tokens→1M tokens+281.5% | artificialanalysis.ai | claims → | |
| Qwen3.6 Max Preview | Qwen | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| Sonar | Perplexity AI | 127.1K tokens→127K tokens−0.1% | artificialanalysis.ai | claims → | |
| Mistral Large | Mistral AI | 128K tokens→32.8K tokens−74.4% | artificialanalysis.ai | claims → | |
| GLM 4.5 Air | Z.ai (Zhipu AI) | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| LongCat 2.0 | Meituan | 1.05M tokens→1M tokens−4.6% | artificialanalysis.ai | claims → | |
| GLM 5.1 | Z.ai (Zhipu AI) | 204.8K tokens→200K tokens−2.3% | artificialanalysis.ai | claims → | |
| Mistral Medium 3 | Mistral AI | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| Qwen3 VL 30B A3B Instruct | Qwen | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| gpt-5.1 | OpenAI | 400K tokens→272K tokens−32.0% | artificialanalysis.ai | claims → | |
| R1 Distill Llama 70B | DeepSeek | 8.19K tokens→128K tokens+1462.5% | artificialanalysis.ai | claims → | |
| Qwen3 14B | Qwen | 131.1K tokens→32.8K tokens−75.0% | artificialanalysis.ai | claims → | |
| Kimi K2 0711 | Moonshot AI | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| Ling 3.0 Flash VL | inclusionAI | 131.1K tokens→262.1K tokens+100.0% | artificialanalysis.ai | claims → | |
| GLM 5 | Z.ai (Zhipu AI) | 204.8K tokens→200K tokens−2.3% | artificialanalysis.ai | claims → | |
| Qwen3 VL 8B Instruct | Qwen | 262.1K tokens→256K tokens−2.3% | artificialanalysis.ai | claims → | |
| gpt-4.1 | OpenAI | 1.05M tokens→1M tokens−4.5% | artificialanalysis.ai | claims → | |
| GLM 4.5V | Z.ai (Zhipu AI) | 65.5K tokens→64K tokens−2.3% | artificialanalysis.ai | claims → | |
| Gemma 3 12B | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | ||
| Gemma 3 27B | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | ||
| Nemotron 3.5 Lightning | NVIDIA | 262.1K tokens→1M tokens+281.5% | artificialanalysis.ai | claims → | |
| MiniMax M3 | MiniMax | 1.05M tokens→1M tokens−4.6% | artificialanalysis.ai | claims → | |
| GLM 5V Turbo | Z.ai (Zhipu AI) | 202.8K tokens→200K tokens−1.4% | artificialanalysis.ai | claims → | |
| Nemotron 3 Nano 30B A3B | NVIDIA | 262.1K tokens→1M tokens+281.5% | artificialanalysis.ai | claims → | |
| MiMo-V2.5 | Xiaomi | 1.05M tokens→1M tokens−4.8% | artificialanalysis.ai | claims → | |
| Gemma 3 4B | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | ||
| Granite 4.0 Micro | IBM | 131K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| gpt-5.5 | OpenAI | 1.05M tokens→922K tokens−12.2% | artificialanalysis.ai | claims → | |
| Sonar Reasoning Pro | Perplexity AI | 128K tokens→127K tokens−0.8% | artificialanalysis.ai | claims → | |
| Hermes 3 70B Instruct | Nous Research | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| GLM 4.5 | Z.ai (Zhipu AI) | 131.1K tokens→128K tokens−2.3% | artificialanalysis.ai | claims → | |
| Solar Pro 4 | Upstage | 524.3K tokens→512K tokens−2.3% | artificialanalysis.ai | claims → | |
| Grok 4.20 | xAI | 2M tokens→1M tokens−50.0% | docs.x.ai | claims → | |
| Grok 4.20 Multi-Agent | xAI | 2M tokens→1M tokens−50.0% | docs.x.ai | claims → | |
| Qwen2.5 | Qwen Team | 128K tokens→131.1K tokens+2.4% | qwenlm.github.io | claims → | |
| Qwen3 235B A22B | Qwen | 131.1K tokens→128K tokens−2.3% | qwenlm.github.io | claims → | |
| qwen3-coder-480b-a35b-instruct | Alibaba Group | 262.1K tokens→256K tokens−2.3% | qwenlm.github.io | claims → | |
| Llama 3.1 70B Instruct | Meta AI | 16.4K tokens→8.19K tokens−50.0% | openrouter.ai | claims → | |
| MiniMax M2.5 | MiniMax | 128K tokens→131.1K tokens+2.4% | openrouter.ai | claims → | |
| Qwen3 235B A22B Instruct 2507 | Qwen | 16.4K tokens→235.9K tokens+1340.0% | openrouter.ai | claims → | |
| Qwen2.5 | Qwen Team | 8K tokens→8.19K tokens+2.4% | qwenlm.github.io | claims → |
New benchmark leaders
New benchmark leaders 15
Benchmarks whose primary-group leader (computed from results observed by each date) differs between 5 Sept 2026 and 12 Sept 2026.
| Benchmark | Leader at 5 Sept 2026 | Score | Leader at 12 Sept 2026 | Score | Metric |
|---|---|---|---|---|---|
| Aider polyglotcoding | no result observed yet | — | gpt-5 (high) | 88 | pass_rate_2official-benchmark |
| Artificial Analysis Intelligence Indexcomposite | no result observed yet | — | Claude Fable 5.1 | 53.37 | indexindependent-evaluator |
| GPQAreasoning | no result observed yet | — | gpt-6-astra-xhigh | 96.26 | accuracyindependent-evaluator |
| Humanity's Last Examknowledge | no result observed yet | — | Claude Fable 5.1 | 59.13 | accuracyindependent-evaluator |
| IFBenchinstruction-following | no result observed yet | — | grok-4-3-medium | 83.33 | accuracyindependent-evaluator |
| LiveBenchgeneral | no result observed yet | — | Claude Fable 5.1 Max Effort | 83.41 | global_averageofficial-benchmark |
| MMMU-Promultimodal | no result observed yet | — | gpt-6-astra | 86.88 | accuracyindependent-evaluator |
| SWE-bench (full test split)coding | no result observed yet | — | claude-3-opus | 3.79 | resolvedofficial-benchmark |
| SWE-bench Litecoding | no result observed yet | — | claude-3-opus | 4.33 | resolvedofficial-benchmark |
| SWE-bench Multilingualcoding | no result observed yet | — | gemini-3-flash | 72.7 | resolvedofficial-benchmark |
| SWE-bench Multimodalcoding | no result observed yet | — | o3 | 35.98 | resolvedofficial-benchmark |
| SWE-bench Verifiedcoding | no result observed yet | — | Claude Opus 4.5 | 76.8 | resolvedofficial-benchmark |
| SciCodecoding | no result observed yet | — | Claude Fable 5.1 | 63.08 | accuracyindependent-evaluator |
| Terminal-Benchagentic | no result observed yet | — | gpt-5-6-sol | 65.91 | accuracyindependent-evaluator |
| τ²-benchagentic | no result observed yet | — | Z.ai GLM 5.2 | 99.12 | pass^1independent-evaluator |
New papers
New papers 32
Provider changes
Provider changes 649
Showing the first 200 of 649 price events. Narrow the scope or the dates to see everything.
Hardware changes
Hardware changes 19
New hardware: MacBook Air (Apple M5) (Apple)
apple_specsNew hardware: MacBook Pro (Apple M5 Max) (Apple)
apple_specsNew hardware: MacBook Pro (Apple M5) (Apple)
apple_specsNew hardware: MacBook Pro (Apple M5 Pro) (Apple)
apple_specsNew hardware: Mac mini (Apple M5 Pro) (Apple)
apple_specsNew hardware: Mac Studio (Apple M5 Max) (Apple)
apple_specsNew hardware: Mac Studio (Apple M5 Ultra) (Apple)
apple_specsNVIDIA DGX Spark: architecture changed from NVIDIA Grace Blackwell to Grace Blackwell (GB10)
ArchitectureNVIDIA Grace Blackwell→Grace Blackwell (GB10)registryNVIDIA DGX Spark: architecture changed from Grace Blackwell (GB10) to NVIDIA Grace Blackwell
ArchitectureGrace Blackwell (GB10)→NVIDIA Grace Blackwellnvidia_specs
Other property changes
Other property changes 200
Openness, licence, status, parameters, release-date and other material property changes (context changes are listed above).
Llama-3.2-1B: openness changed from open-weights to restricted
Opennessopen-weights→restrictedhuggingfaceLlama 4 Scout: openness changed from open-weights to restricted
Opennessopen-weights→restrictedhuggingfaceLlama-3.2-3B: openness changed from open-weights to restricted
Opennessopen-weights→restrictedhuggingfaceLlama 3.1 8B: openness changed from open-weights to restricted
Opennessopen-weights→restrictedhuggingfaceLlama-3.1-405B: openness changed from open-weights to restricted
Opennessopen-weights→restrictedhuggingfaceGemma 3 4B: openness changed from open-weights to restricted
Opennessopen-weights→restrictedhuggingfacetiny-aya-global: openness changed from open-weights to restricted
Opennessopen-weights→restrictedhuggingfaceLlama-3.2-3B: openness changed from restricted to open-weights
Opennessrestricted→open-weightsartificial_analysisLlama-3.2-1B: openness changed from restricted to open-weights
Opennessrestricted→open-weightsartificial_analysisLlama-3.1-405B: openness changed from restricted to open-weights
Opennessrestricted→open-weightsartificial_analysisLlama 4 Scout: openness changed from restricted to open-weights
Opennessrestricted→open-weightsartificial_analysisGemma 3 4B: openness changed from restricted to open-weights
Opennessrestricted→open-weightsartificial_analysisLlama 3.1 8B: openness changed from restricted to open-weights
Opennessrestricted→open-weightsartificial_analysistiny-aya-global: openness changed from restricted to open-weights
Opennessrestricted→open-weightsartificial_analysisdeepseek-r1: openness changed from open-weights to open-source
Opennessopen-weights→open-sourcedeepseekDeepSeek V3 0324: openness changed from open-weights to open-source
Opennessopen-weights→open-sourcedeepseekDeepSeek V3.2 Exp: openness changed from open-weights to open-source
Opennessopen-weights→open-sourcedeepseekdeepseek-v4-pro: openness changed from open-weights to proprietary
Opennessopen-weights→proprietarydeepseekDeepSeek V4 Flash Vision Exp: openness changed from open-weights to proprietary
Opennessopen-weights→proprietarydeepseekQwen3 235B A22B: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterphoenix: latest version changed from arize-phoenix: v20.10.0 to arize-phoenix: v20.11.0
Latest versionarize-phoenix: v20.10.0→arize-phoenix: v20.11.0githubpydantic-ai: latest version changed from v2.42.0 (2026-09-08) to v2.43.0 (2026-09-11)
Latest versionv2.42.0 (2026-09-08)→v2.43.0 (2026-09-11)githubgemini-cli: latest version changed from Release v0.61.0-nightly.20260911.ged2ac40df to Release v0.61.0-nightly.20260912.g9c1b0a610
Latest versionRelease v0.61.0-nightly.20260911.ged2ac40df→Release v0.61.0-nightly.20260912.g9c1b0a610githubllama.cpp: latest version changed from b10917 to b10919
Latest versionb10917→b10919githubpytorch: latest version changed from viable/strict/1789165518 to trunk/363267c09a72be003607d9e0e50df93150e4cb9a: [torchcom…
Latest versionviable/strict/1789165518→trunk/363267c09a72be003607d9e0e50df93150e4cb9a: [torchcomms hash update] update the pinned torchcommgithubGemma 3 12B: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterGLM 4.5: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterKimi K2 0711: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterQwen3 235B A22B Instruct 2507: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterGemma 3 27B: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterSonar Reasoning Pro: context length changed from 127000 to 128000
Context window127K tokens→128K tokensopenrouterGLM 4.5 Air: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterDeepSeek V3.1: context length changed from 128000 to 163840
Context window128K tokens→163.8K tokensopenrouterDeepSeek V3 0324: context length changed from 128000 to 163840
Context window128K tokens→163.8K tokensopenrouterDeepSeek V3: context length changed from 128000 to 163840
Context window128K tokens→163.8K tokensopenrouterGPT-3.5 Turbo: context length changed from 4096 to 16385
Context window4.1K tokens→16.4K tokensopenrouterGemma 3 4B: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterDeepSeek V3.2 Exp: context length changed from 128000 to 163840
Context window128K tokens→163.8K tokensopenrouterMistral Medium 3: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterQwen3 14B: context length changed from 32768 to 131072
Context window32.8K tokens→131.1K tokensopenrouterPhi 4: context length changed from 16000 to 16384
Context window16K tokens→16.4K tokensopenrouterKimi K2.6: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterStep 3.5 Flash: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenroutergpt-4.1: context length changed from 1000000 to 1047576
Context window1M tokens→1.05M tokensopenrouterQwen3 Coder Next: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenroutergpt-4.1-nano: context length changed from 1000000 to 1047576
Context window1M tokens→1.05M tokensopenrouterLing 3.0 Flash VL: context length changed from 262144 to 131072
Context window262.1K tokens→131.1K tokensopenroutergpt-5.1: context length changed from 272000 to 400000
Context window272K tokens→400K tokensopenrouterQwen3 32B: context length changed from 32768 to 131072
Context window32.8K tokens→131.1K tokensopenrouterMuse Spark 1.3: context length changed from 1000000 to 1048576
Context window1M tokens→1.05M tokensopenrouterDeepSeek V3.1 Terminus: context length changed from 128000 to 163840
Context window128K tokens→163.8K tokensopenrouterLlama 4 Maverick: context length changed from 1000000 to 1048576
Context window1M tokens→1.05M tokensopenrouterHermes 4 405B: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterQwen3 VL 30B A3B Instruct: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterKimi K2.5: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterKimi K2.7 Code: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterMistral Saba: context length changed from 32000 to 32768
Context window32K tokens→32.8K tokensopenrouterReka Flash 3: context length changed from 128000 to 65536
Context window128K tokens→65.5K tokensopenrouterGLM 5.3: context length changed from 1000000 to 1310720
Context window1M tokens→1.31M tokensopenrouterGLM 5.3 Flash: context length changed from 1000000 to 1310720
Context window1M tokens→1.31M tokensopenrouterMiniMax M3: context length changed from 1000000 to 1048576
Context window1M tokens→1.05M tokensopenrouterSolar Pro 3: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterSolar Pro 4: context length changed from 512000 to 524288
Context window512K tokens→524.3K tokensopenrouterQwen3.6 Max Preview: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterGemma 4 26B A4B: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterGLM 5.1: context length changed from 200000 to 204800
Context window200K tokens→204.8K tokensopenroutergpt-5.5: context length changed from 922000 to 1050000
Context window922K tokens→1.05M tokensopenrouterGemma 4 31B: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterLlama 4 Scout: context length changed from 10000000 to 1310720
Context window10M tokens→1.31M tokensopenrouterGLM 5 Turbo: context length changed from 200000 to 202752
Context window200K tokens→202.8K tokensopenrouterMiMo-V2.5: context length changed from 1000000 to 1050000
Context window1M tokens→1.05M tokensopenrouterLongCat 2.0: context length changed from 1000000 to 1048756
Context window1M tokens→1.05M tokensopenrouterMistral Large 2.0: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterKAT-Coder-Pro V2: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenroutergpt-4.1-mini: context length changed from 1000000 to 1047576
Context window1M tokens→1.05M tokensopenrouterMiMo-V2.5-Pro: context length changed from 1000000 to 1050000
Context window1M tokens→1.05M tokensopenrouterSonar: context length changed from 127000 to 127072
Context window127K tokens→127.1K tokensopenroutergpt-5.5-pro: context length changed from 922000 to 1050000
Context window922K tokens→1.05M tokensopenrouterMistral Large: context length changed from 32768 to 128000
Context window32.8K tokens→128K tokensopenrouterQwen3.8 2.4T A95B: context length changed from 983616 to 1048576
Context window983.6K tokens→1.05M tokensopenrouterQwen3.8 27B: context length changed from 256000 to 1000000
Context window256K tokens→1M tokensopenrouterQwen3.8 Flash: context length changed from 256000 to 1000000
Context window256K tokens→1M tokensopenrouterGLM 4.5V: context length changed from 64000 to 65536
Context window64K tokens→65.5K tokensopenrouterNemotron 3.5 Lightning: context length changed from 1000000 to 262144
Context window1M tokens→262.1K tokensopenrouterInkling Small: context length changed from 1000000 to 1048576
Context window1M tokens→1.05M tokensopenrouterHy3: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterNemotron 3 Super: context length changed from 1000000 to 262144
Context window1M tokens→262.1K tokensopenrouterHy3 preview: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterGLM 4.6V: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterGLM 4.7: context length changed from 200000 to 204800
Context window200K tokens→204.8K tokensopenrouterDeepSeek V3.2: context length changed from 128000 to 163840
Context window128K tokens→163.8K tokensopenrouterR1 Distill Llama 70B: context length changed from 128000 to 8192
Context window128K tokens→8.19K tokensopenrouterTrinity Large Thinking: context length changed from 512000 to 262144
Context window512K tokens→262.1K tokensopenrouterGLM 5V Turbo: context length changed from 200000 to 202752
Context window200K tokens→202.8K tokensopenrouterGLM 5: context length changed from 200000 to 204800
Context window200K tokens→204.8K tokensopenrouterQwen3 VL 8B Instruct: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterQwen3 VL 32B Instruct: context length changed from 256000 to 131072
Context window256K tokens→131.1K tokensopenrouterKimi K2 Thinking: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterHermes 3 70B Instruct: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterNemotron 3 Nano 30B A3B: context length changed from 1000000 to 262144
Context window1M tokens→262.1K tokensopenrouterGranite 4.0 Micro: context length changed from 128000 to 131000
Context window128K tokens→131K tokensopenroutergpt-3.5-turbo: context length changed from 4096 to 4095
Context window4.1K tokens→4.1K tokensopenroutergpt-4: context length changed from 8192 to 8191
Context window8.19K tokens→8.19K tokensopenrouterGLM 4.6: context length changed from 200000 to 204800
Context window200K tokens→204.8K tokensopenrouterMistral Medium 3.1: context length changed from 128000 to 131072
Context window128K tokens→131.1K tokensopenrouterMistral Small 3: context length changed from 32000 to 32768
Context window32K tokens→32.8K tokensopenrouterQwen2.5 Coder 32B Instruct: context length changed from 131072 to 32768
Context window131.1K tokens→32.8K tokensopenrouterKimi K2 0905: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterDevstral 2: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterQwen3 Max Thinking: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterInkling: context length changed from 1000000 to 1048576
Context window1M tokens→1.05M tokensopenrouterpytorch: latest version changed from viable/strict/1789158535: [ROCm][CD] Add gfx1250 (MI450) … to viable/strict/1789165518
Latest versionviable/strict/1789158535: [ROCm][CD] Add gfx1250 (MI450) to the nightly wheel arch list (#196610)→viable/strict/1789165518githubgpt-3.5-turbo: context length changed from 4095 to 4096
Context window4.1K tokens→4.1K tokensartificial_analysisDeepSeek V3: context length changed from 163840 to 128000
Context window163.8K tokens→128K tokensartificial_analysisHermes 4 405B: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisGemma 4 26B A4B: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisSolar Pro 3: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisGLM 5.1: context length changed from 204800 to 200000
Context window204.8K tokens→200K tokensartificial_analysisQwen3 VL 32B Instruct: context length changed from 131072 to 256000
Context window131.1K tokens→256K tokensartificial_analysisKimi K2 0905: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisNemotron 3 Super: context length changed from 262144 to 1000000
Context window262.1K tokens→1M tokensartificial_analysisQwen3.8 27B: context length changed from 1000000 to 256000
Context window1M tokens→256K tokensartificial_analysisGemma 4 31B: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisgpt-4.1-nano: context length changed from 1047576 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisKimi K2 Thinking: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisStep 3.5 Flash: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisMistral Large: context length changed from 128000 to 32768
Context window128K tokens→32.8K tokensartificial_analysisgpt-4: context length changed from 8191 to 8192
Context window8.19K tokens→8.19K tokensartificial_analysisSonar: context length changed from 127072 to 127000
Context window127.1K tokens→127K tokensartificial_analysisKimi K2.7 Code: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisGLM 4.7: context length changed from 204800 to 200000
Context window204.8K tokens→200K tokensartificial_analysisQwen3.6 Max Preview: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisHy3 preview: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisR1 Distill Llama 70B: context length changed from 8192 to 128000
Context window8.19K tokens→128K tokensartificial_analysisQwen3 32B: context length changed from 131072 to 32768
Context window131.1K tokens→32.8K tokensartificial_analysisKimi K2 0711: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisQwen3.8 2.4T A95B: context length changed from 1048576 to 983616
Context window1.05M tokens→983.6K tokensartificial_analysisLongCat 2.0: context length changed from 1048756 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisLing 3.0 Flash VL: context length changed from 131072 to 262144
Context window131.1K tokens→262.1K tokensartificial_analysisGLM 5.3: context length changed from 1310720 to 1000000
Context window1.31M tokens→1M tokensartificial_analysisMistral Medium 3: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisGPT-3.5 Turbo: context length changed from 16385 to 4096
Context window16.4K tokens→4.1K tokensartificial_analysisInkling Small: context length changed from 1048576 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisGemma 3 27B: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisGLM 4.5: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisDeepSeek V3 0324: context length changed from 163840 to 128000
Context window163.8K tokens→128K tokensartificial_analysisGLM 4.5 Air: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisKimi K2.6: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisgpt-4.1: context length changed from 1047576 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisSolar Pro 4: context length changed from 524288 to 512000
Context window524.3K tokens→512K tokensartificial_analysisReka Flash 3: context length changed from 65536 to 128000
Context window65.5K tokens→128K tokensartificial_analysisMistral Medium 3.1: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisgpt-5.1: context length changed from 400000 to 272000
Context window400K tokens→272K tokensartificial_analysisQwen3 VL 30B A3B Instruct: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisGLM 5 Turbo: context length changed from 202752 to 200000
Context window202.8K tokens→200K tokensartificial_analysisGLM 4.6: context length changed from 204800 to 200000
Context window204.8K tokens→200K tokensartificial_analysisMistral Saba: context length changed from 32768 to 32000
Context window32.8K tokens→32K tokensartificial_analysisgpt-5.5-pro: context length changed from 1050000 to 922000
Context window1.05M tokens→922K tokensartificial_analysisQwen2.5 Coder 32B Instruct: context length changed from 32768 to 131072
Context window32.8K tokens→131.1K tokensartificial_analysisDevstral 2: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisGemma 3 4B: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisQwen3.8 Flash: context length changed from 1000000 to 256000
Context window1M tokens→256K tokensartificial_analysisgpt-4.1-mini: context length changed from 1047576 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisMistral Large 2.0: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisDeepSeek V3.1 Terminus: context length changed from 163840 to 128000
Context window163.8K tokens→128K tokensartificial_analysisDeepSeek V3.2: context length changed from 163840 to 128000
Context window163.8K tokens→128K tokensartificial_analysisMuse Spark 1.3: context length changed from 1048576 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisGLM 5V Turbo: context length changed from 202752 to 200000
Context window202.8K tokens→200K tokensartificial_analysisKimi K2.5: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisInkling: context length changed from 1048576 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisMiMo-V2.5-Pro: context length changed from 1050000 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisQwen3 235B A22B Instruct 2507: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisMistral Small 3: context length changed from 32768 to 32000
Context window32.8K tokens→32K tokensartificial_analysisGLM 4.6V: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisSonar Reasoning Pro: context length changed from 128000 to 127000
Context window128K tokens→127K tokensartificial_analysisDeepSeek V3.2 Exp: context length changed from 163840 to 128000
Context window163.8K tokens→128K tokensartificial_analysisGemma 3 12B: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisHermes 3 70B Instruct: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisMiniMax M3: context length changed from 1048576 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisNemotron 3 Nano 30B A3B: context length changed from 262144 to 1000000
Context window262.1K tokens→1M tokensartificial_analysisGranite 4.0 Micro: context length changed from 131000 to 128000
Context window131K tokens→128K tokensartificial_analysisDeepSeek V3.1: context length changed from 163840 to 128000
Context window163.8K tokens→128K tokensartificial_analysisGLM 5: context length changed from 204800 to 200000
Context window204.8K tokens→200K tokensartificial_analysisPhi 4: context length changed from 16384 to 16000
Context window16.4K tokens→16K tokensartificial_analysisGLM 5.3 Flash: context length changed from 1310720 to 1000000
Context window1.31M tokens→1M tokensartificial_analysisLlama 4 Scout: context length changed from 1310720 to 10000000
Context window1.31M tokens→10M tokensartificial_analysisQwen3 Coder Next: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisTrinity Large Thinking: context length changed from 262144 to 512000
Context window262.1K tokens→512K tokensartificial_analysisHy3: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisQwen3 VL 8B Instruct: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisNemotron 3.5 Lightning: context length changed from 262144 to 1000000
Context window262.1K tokens→1M tokensartificial_analysisKAT-Coder-Pro V2: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisMiMo-V2.5: context length changed from 1050000 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisQwen3 Max Thinking: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisQwen3 14B: context length changed from 131072 to 32768
Context window131.1K tokens→32.8K tokensartificial_analysisGLM 4.5V: context length changed from 65536 to 64000
Context window65.5K tokens→64K tokensartificial_analysisgpt-5.5: context length changed from 1050000 to 922000
Context window1.05M tokens→922K tokensartificial_analysisLlama 4 Maverick: context length changed from 1048576 to 1000000
Context window1.05M tokens→1M tokensartificial_analysisGrok 4.20: context length changed from 2000000 to 1000000
Context window2M tokens→1M tokensxaiGrok 4.20 Multi-Agent: context length changed from 2000000 to 1000000
Context window2M tokens→1M tokensxai
Other new entities
Other new entities 95
| Entity | Type | Organization | First seen |
|---|---|---|---|
| SWE-Bench Pro | Benchmark | — | 12 Sept 2026 |
| MMTEB | Benchmark | — | 12 Sept 2026 |
| Terminal-Bench 2.0 | Benchmark | — | 12 Sept 2026 |
| ARC-AGI-2 | Benchmark | — | 12 Sept 2026 |
| AIME 2024 | Benchmark | — | 12 Sept 2026 |
| LiveBench Instruction Following | Benchmark | — | 12 Sept 2026 |
| LiveBench Language | Benchmark | — | 12 Sept 2026 |
| LiveBench Data Analysis | Benchmark | — | 12 Sept 2026 |
| LiveBench Mathematics | Benchmark | — | 12 Sept 2026 |
| LiveBench Agentic Coding | Benchmark | — | 12 Sept 2026 |
| LiveBench Coding | Benchmark | — | 12 Sept 2026 |
| LiveBench Reasoning | Benchmark | — | 12 Sept 2026 |
| Aider polyglot — well-formed responses | Benchmark | — | 12 Sept 2026 |
| GPQA Diamond | Benchmark | — | 12 Sept 2026 |
| LG AI Research | Lab | — | 12 Sept 2026 |
| Kuaishou | Company | — | 12 Sept 2026 |
| Qwen API | Provider | — | 12 Sept 2026 |
| Mistral AI | Provider | — | 12 Sept 2026 |
| Skild AI | Company | — | 12 Sept 2026 |
| Vittoria Cavicchioli | Researcher | — | 12 Sept 2026 |
| Michele Pestarino | Researcher | — | 12 Sept 2026 |
| Davide Malvezzi | Researcher | — | 12 Sept 2026 |
| Matt White | Researcher | — | 12 Sept 2026 |
| Yiyun Su | Researcher | — | 12 Sept 2026 |
| Zelin Li | Researcher | — | 12 Sept 2026 |
| Daniel Kirchner | Researcher | — | 12 Sept 2026 |
| Christoph Benzmueller | Researcher | — | 12 Sept 2026 |
| Christopher Buccafusco | Researcher | — | 12 Sept 2026 |
| Emily Wenger | Researcher | — | 12 Sept 2026 |
| Nirav Patel | Researcher | — | 12 Sept 2026 |
| Paul Lerner | Researcher | — | 12 Sept 2026 |
| Salim Hafid | Researcher | — | 12 Sept 2026 |
| Pierre-Antoine Lequeu | Researcher | — | 12 Sept 2026 |
| Hucheng Yang | Researcher | — | 12 Sept 2026 |
| Lang An | Researcher | — | 12 Sept 2026 |
| Yixiong Xiao | Researcher | — | 12 Sept 2026 |
| Hang Lyu | Researcher | — | 12 Sept 2026 |
| Matteo Vaccargiu | Researcher | — | 12 Sept 2026 |
| Daniel Graziotin | Researcher | — | 12 Sept 2026 |
| Giuseppe Destefanis | Researcher | — | 12 Sept 2026 |
| Cato Elia Kurtz | Researcher | — | 12 Sept 2026 |
| Blake G. Fitch | Researcher | — | 12 Sept 2026 |
| Weiran Wang | Researcher | — | 12 Sept 2026 |
| Zhouyuan Huo | Researcher | — | 12 Sept 2026 |
| Aaron Isidore Grace | Researcher | — | 12 Sept 2026 |
| Zhangkai Wu | Researcher | — | 12 Sept 2026 |
| Jiaqi Zhang | Researcher | — | 12 Sept 2026 |
| Yiqi Wang | Researcher | — | 12 Sept 2026 |
| Benedikt Bollig | Researcher | — | 12 Sept 2026 |
| Ronghe Wang | Researcher | — | 12 Sept 2026 |
| Priyam Srivastava | Researcher | — | 12 Sept 2026 |
| Xin Jin | Researcher | — | 12 Sept 2026 |
| Yuhang Gu | Researcher | — | 12 Sept 2026 |
| Chengtao Lai | Researcher | — | 12 Sept 2026 |
| Zhongchun Zhou | Researcher | — | 12 Sept 2026 |
| Beatriz Navarro Lameda | Researcher | — | 12 Sept 2026 |
| Nikoleta Kalaydzhieva | Researcher | — | 12 Sept 2026 |
| Benjamin J. Walker | Researcher | — | 12 Sept 2026 |
| Sebastian Szyller | Researcher | — | 12 Sept 2026 |
| Vasisht Duddu | Researcher | — | 12 Sept 2026 |
| Asim Waheed | Researcher | — | 12 Sept 2026 |
| Hridoy Sankar Dutta | Researcher | — | 12 Sept 2026 |
| Keshav Sood | Researcher | — | 12 Sept 2026 |
| Kamel Kamel | Researcher | — | 12 Sept 2026 |
| Yinghao Tang | Researcher | — | 12 Sept 2026 |
| Luoxuan Weng | Researcher | — | 12 Sept 2026 |
| Zhen Wen | Researcher | — | 12 Sept 2026 |
| Tommy Sha | Researcher | — | 12 Sept 2026 |
| Stella Zhao | Researcher | — | 12 Sept 2026 |
| Baokun Wang | Researcher | — | 12 Sept 2026 |
| Jinyong Wen | Researcher | — | 12 Sept 2026 |
| Jinsong Shu | Researcher | — | 12 Sept 2026 |
| Stefanie Rinderle-Ma | Researcher | — | 12 Sept 2026 |
| Michel Kunkler | Researcher | — | 12 Sept 2026 |
| Matthias Endres | Researcher | — | 12 Sept 2026 |
| Miles Tidmarsh | Researcher | — | 12 Sept 2026 |
| Jasmine Brazilek | Researcher | — | 12 Sept 2026 |
| Can Li | Researcher | — | 12 Sept 2026 |
| Qinzheng Wang | Researcher | — | 12 Sept 2026 |
| Qi Liu | Researcher | — | 12 Sept 2026 |
| Avi Sharma | Researcher | — | 12 Sept 2026 |
| Shrenil Shaun Sharma | Researcher | — | 12 Sept 2026 |
| Lei Shen | Researcher | — | 12 Sept 2026 |
| Huize Yu | Researcher | — | 12 Sept 2026 |
| Mingkang Liu | Researcher | — | 12 Sept 2026 |
| Roberto Garrone | Researcher | — | 12 Sept 2026 |
| Patrick Cousot | Researcher | — | 12 Sept 2026 |
| Jade Alglave | Researcher | — | 12 Sept 2026 |
| Yuyang He | Researcher | — | 12 Sept 2026 |
| Xiye Ma | Researcher | — | 12 Sept 2026 |
| Wanting Wang | Researcher | — | 12 Sept 2026 |
| H. D. Lethe jr | Researcher | — | 12 Sept 2026 |
| S. F. M. van Vlijmen | Researcher | — | 12 Sept 2026 |
| Zexi Liu | Researcher | — | 12 Sept 2026 |
| Yuzhu Cai | Researcher | — | 12 Sept 2026 |
Gone entities
Gone entities 0
No entity disappeared between these dates
“Gone” means an entity present at the first date is no longer current at the second (merged or retired) — its record and history are kept. Events are keyed on occurred_at; back-filled history is excluded (tick “include backfill” to add it). Per-entity history: any entity's History tab. Atlas as of 5 Sept 2026 · Methodology →