gpt-oss-20b
OpenAIdevelopers.openai.com/api/docs/models/gpt-oss-20
Medium-sized open-weight model for low latency
Updated 29 min ago · first seen 11 Sept 2026
model_01M293VC6B06HH77JHPAVH04D2
- Parameters
- 20.9B
- T2 · 3 h ago
- Context
- 131.1K tokens
- T2 · 3 h ago
- Released
- 4 Aug 2025
- T2 · 29 min ago
- License
- apache-2.0
- T2 · 2 h ago
- Knowledge cutoff
- Jun 2024
- T2 · 3 h ago
Specification
- Release date
- 4 Aug 2025
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 29 min agomedium
- Openness
- open-weights
Source:OpenAI Platform docsT1observed 3 h agohigh
- License
- apache-2.0
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Architecture
- GptOssForCausalLM
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Parameters
- 20.9B
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 3 h agomedium
- Context window
- 131.1K tokens
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Max output
- 118K tokens
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Knowledge cutoff
- Jun 2024
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Modalities
- text
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Input modalities
- text
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Output modalities
- text
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Tokenizer
- GPT
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- API model id
- gpt-oss-20b
Source:OpenAI Platform docsT1observed 3 h agohigh
- File size
- 13.8 GB
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Hugging Face repo
- openai/gpt-oss-20b
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Pipeline tag
- text-generation
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 3 h agomedium
- Official page
Source:OpenAI Platform docsT1observed 3 h agohigh
- Model card
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 3 h agomedium
- Downloads
- 6,599,829
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 29 min agomedium
- Likes
- 5,019
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 29 min agomedium
- Gated
- No
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Groq model id
- openai/gpt-oss-20b
Source:Groq — docs & pricingT1observed 3 h agohigh
- Hf inference providers
- deepinfra, featherless-ai, groq, novita, nscale, ovhcloud, together
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Last modified
- 2025-08-26T17:25:47+00:00
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 3 h agomedium
- Library name
- transformers
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Aa median output tokens per second
- 179.4
Source:Artificial AnalysisT2observed 1 h agomedium
- Downloads all time
- 96,531,522
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 29 min agomedium
- Model type
- gpt_oss
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Openrouter id
- openai/gpt-oss-20b
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Supported parameters
- frequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Tags
- transformers, safetensors, gpt_oss, text-generation, vllm, 8-bit, mxfp4
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Weights dtype
- BF16, U8
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →
Provenance
Attributed facts
36
Source tiers
T1T25 / 31
Freshest observation
29 min ago
Conflicts
None
Modalities
- Modalities
- text
- Input
- text
- Output
- text
Capabilities
Tool calling
Yes
OpenRouter public model & pricing listing · T2
Structured output
Yes
OpenRouter public model & pricing listing · T2
Reasoning
Yes
OpenRouter public model & pricing listing · T2
Vision
Unavailable
Audio
Unavailable
Fine-tuning available
Unavailable
- Context window
- 131.1K tokens
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Max output
- 118K tokens
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Knowledge cutoff
- Jun 2024
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
- Tokenizer
- GPT
Source:OpenRouter public model & pricing listingT2observed 3 h agomedium
| Benchmark | Score | Metric | Config | Evaluated | Source |
|---|---|---|---|---|---|
| τ²-bench | 60.23% | pass^1 | aa_slug=gpt-oss-20b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Artificial Analysis Intelligence Index | 9.04 | index | aa_slug=gpt-oss-20b · version=4.3 · estimated=false | — | artificialanalysis.aiT2 |
| GPQA | 68.79% | accuracy | aa_slug=gpt-oss-20b · variant=GPQA Diamond · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Humanity's Last Exam | 10.98% | accuracy | aa_slug=gpt-oss-20b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| IFBench | 65.1% | accuracy | aa_slug=gpt-oss-20b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| SciCode | 38.89% | accuracy | aa_slug=gpt-oss-20b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Terminal-Bench | 10.61% | accuracy | aa_slug=gpt-oss-20b · variant=hard · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Terminal-Bench | 13.86% | accuracy | aa_slug=gpt-oss-20b · variant=v2.1 · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Terminal-Bench | 0% | accuracy | aa_slug=gpt-oss-20b · variant=v4.0 · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
Scores are reported as published, with their evaluation configuration (harness, prompting, judge). Results with different configs are not directly comparable — see methodology.
Current prices
| Provider | Input / 1M | Output / 1M | Cached in | Batch in / out | Context | Observed | Source |
|---|---|---|---|---|---|---|---|
| OpenAI APIopenai/gpt-oss-20b | $0.03 | $0.13 | $0.03 | — | 131.1K | 1 h ago | openrouter.aiT2 |
| OpenAI APIopenai/gpt-oss-20b:batch | $0.05 | $0.20 | — | — | 131.1K | 1 h ago | openrouter.aiT2 |
| GroqCloudopenai/gpt-oss-20b | $0.075 | $0.30 | — | — | 131.1K | 1 h ago | console.groq.comT1 |
USD per 1M tokens as published by each provider (USD). Rows are append-only: every change is kept in the history below.
Available through 2
Price history
Input price · USD / 1M tokens
Output price · USD / 1M tokens
- OpenAI API
- GroqCloud
Memory need = bytes per parameter (4-bit ≈ 0.5 × 1.15 overhead, 8-bit 1.0, fp16 2.0) + a KV-cache allowance. Not a measurement.
| Hardware | Quantization | Memory | Est. need | Fits |
|---|---|---|---|---|
| NVIDIA DGX B200 | 4bit | 1,440 GB | 12.5 GB | Yes |
| Apple M3 Ultra | 4bit | — | 12.5 GB | Yes |
| Mac Studio (Apple M5 Ultra) | 4bit | — | 12.5 GB | Yes |
| AMD Instinct MI325X | 4bit | 256 GB | 12.5 GB | Yes |
| AMD Instinct MI300X | 4bit | 192 GB | 12.5 GB | Yes |
| Apple M2 Ultra | 4bit | — | 12.5 GB | Yes |
| NVIDIA B200 | 4bit | 180 GB | 12.5 GB | Yes |
| NVIDIA H200 | 4bit | 141 GB | 12.5 GB | Yes |
| NVIDIA H200 NVL | 4bit | 141 GB | 12.5 GB | Yes |
| Apple M1 Ultra | 4bit | — | 12.5 GB | Yes |
| Apple M3 Max | 4bit | — | 12.5 GB | Yes |
| Apple M4 Max | 4bit | — | 12.5 GB | Yes |
| Mac Studio (Apple M5 Max) | 4bit | — | 12.5 GB | Yes |
| MacBook Pro (Apple M5 Max) | 4bit | — | 12.5 GB | Yes |
| NVIDIA DGX Spark | 4bit | 128 GB | 12.5 GB | Yes |
| Apple M2 Max | 4bit | — | 12.5 GB | Yes |
| NVIDIA H100 NVL | 4bit | 94 GB | 12.5 GB | Yes |
| NVIDIA A100 80GB | 4bit | 80 GB | 12.5 GB | Yes |
| NVIDIA H100 SXM | 4bit | 80 GB | 12.5 GB | Yes |
| Apple M1 Max | 4bit | — | 12.5 GB | Yes |
| Apple M4 Pro | 4bit | — | 12.5 GB | Yes |
| Mac mini (Apple M5 Pro) | 4bit | — | 12.5 GB | Yes |
| MacBook Pro (Apple M5 Pro) | 4bit | — | 12.5 GB | Yes |
| Apple M3 Pro | 4bit | — | 12.5 GB | Yes |
| Apple M1 Pro | 4bit | — | 12.5 GB | Yes |
| Apple M2 Pro | 4bit | — | 12.5 GB | Yes |
| Apple M4 | 4bit | — | 12.5 GB | Yes |
| iMac (Apple M4) | 4bit | — | 12.5 GB | Yes |
| Mac mini (Apple M6) | 4bit | — | 12.5 GB | Yes |
| MacBook Air (Apple M5) | 4bit | — | 12.5 GB | Yes |
| MacBook Pro (Apple M5) | 4bit | — | 12.5 GB | Yes |
| NVIDIA GeForce RTX 5090 | 4bit | 32 GB | 12.5 GB | Yes |
| Apple M2 | 4bit | — | 12.5 GB | Yes |
| Apple M3 | 4bit | — | 12.5 GB | Yes |
| NVIDIA GeForce RTX 3090 | 4bit | 24 GB | 12.5 GB | Yes |
| NVIDIA GeForce RTX 4090 | 4bit | 24 GB | 12.5 GB | Yes |
| Apple M1 | 4bit | — | 12.5 GB | Yes |
Ancestors 0
None recorded.
Descendants 1
Papers 1
- arXiv:2508.10925Active35
Repositories 0
No repositories linked yet.
As of
Rewind the record: see this entity's attributes exactly as AI Atlas knew them on a given day.
Claim history · Hf inference providers
Hf inference providershf_inference_providers1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| deepinfra, featherless-ai, groq, novita, nscale, ovhcloud, together | → current | current | Hugging Face Hub (public pages, model cards, papers)T2 | medium | deterministic |
Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →
gpt-oss-20b scores 10.61% on Terminal-Bench
artificial_analysisgpt-oss-20b scores 13.86% on Terminal-Bench
artificial_analysisgpt-oss-20b scores 10.98% on Humanity's Last Exam
artificial_analysisgpt-oss-20b scores 9.04 on Artificial Analysis Intelligence Index
artificial_analysisgpt-oss-20b: release date changed from 2025-08-04 to 2025-08-05
Release date4 Aug 2025→5 Aug 2025openroutergpt-oss-20b: release date changed from 2025-08-05 to 2025-08-04
Release date5 Aug 2025→4 Aug 2025huggingfaceGroqCloud lists gpt-oss-20b at $0.075 in / $0.3 out per 1M tokens
groq_pricingOpenAI API lists gpt-oss-20b at $0.05 in / $0.2 out per 1M tokens
openrouterOpenAI API lists gpt-oss-20b at $0.03 in / $0.13 out per 1M tokens
openrouter
| Source | Document | Type | Tier | Last observed | Snapshots |
|---|---|---|---|---|---|
| OpenAI Platform docs | developers.openai.com/api/docs/models.md | model_docs | T1· Official | 9 min ago | 1 |
| Groq — docs & pricing | console.groq.com/docs/models | model_docs | T1· Official | 1 h ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/openai/gpt-oss-safeguard-20b | model_page | T2· Quality secondary | 28 min ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/openai/gpt-oss-20b | model_page | T2· Quality secondary | 29 min ago | 3 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/mlx-community/gpt-oss-20b-MXFP4-Q8 | model_page | T2· Quality secondary | 33 min ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/models?author=openai&p=0&sort=downloads | listing | T2· Quality secondary | 58 min ago | 3 |
| Artificial Analysis | artificialanalysis.ai/leaderboards/models | leaderboard | T2· Quality secondary | 1 h ago | 1 |
| OpenRouter public model & pricing listing | openrouter.ai/api/v1/models | catalogue | T2· Quality secondary | 1 h ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/openai/gpt-oss-20b/raw/main/README.md | model_card | T2· Quality secondary | 1 h ago | 1 |
Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.