AI Atlas as of 12 Mar 2026
Pick any date: the models that existed, their context lengths and status at the time, the provider prices valid that day, the benchmark leaders known by then, the hardware. Every value opens the claim that establishes it, with its validity interval.
ReconstructedThis date is earlier than AI Atlas's first observation — the state is reconstructed from dated claims and release dates, not from direct observation.
Observation history starts . AI Atlas observation history starts at 2026-09-11T20:50:56.012288+00:00. This date is earlier: the state is RECONSTRUCTED from claims with effective dates and from release dates — not from direct observation.
Models
Models as of 12 Mar 2026 711
| Model | Organization | Params | Context (as of) | Context (today) | Openness | Status (as of) | Status (today) | Released | Basis | History |
|---|---|---|---|---|---|---|---|---|---|---|
| Gemma 4 26B A4B | — | — | 262.1K | — | — | active | 11 Mar 2026 | reconstructed | as of 2026-03-12 → | |
| Gemma 4 31B | — | — | 262.1K | — | — | active | 11 Mar 2026 | reconstructed | as of 2026-03-12 → | |
| Nemotron 3 Super | NVIDIA | — | — | 262.1K | — | — | active | 11 Mar 2026 | reconstructed | as of 2026-03-12 → |
| granite-speech-4.1-2b-nar | IBM | — | — | — | — | — | active | 10 Mar 2026 | reconstructed | as of 2026-03-12 → |
| Seed-2.0-Lite | ByteDance | — | — | 262.1K | — | — | active | 10 Mar 2026 | reconstructed | as of 2026-03-12 → |
| FLUX.2-klein-9b-kv | Black Forest Labs | — | — | — | — | — | active | 9 Mar 2026 | reconstructed | as of 2026-03-12 → |
| gpt-5.4 | OpenAI | — | — | 1.05M | — | — | deprecated | 5 Mar 2026 | reconstructed | as of 2026-03-12 → |
| gpt-5.4-pro | OpenAI | — | — | 1.05M | — | — | deprecated | 5 Mar 2026 | reconstructed | as of 2026-03-12 → |
| Mercury 2 | Inception | — | — | 128K | — | — | active | 4 Mar 2026 | reconstructed | as of 2026-03-12 → |
| Gemini 3.1 Flash-Lite Preview | — | — | 1.05M | — | — | retired | 3 Mar 2026 | reconstructed | as of 2026-03-12 → | |
| gemma-4-E2B-it | — | — | — | — | — | active | 2 Mar 2026 | reconstructed | as of 2026-03-12 → | |
| gemma-4-E4B-it | — | — | — | — | — | active | 2 Mar 2026 | reconstructed | as of 2026-03-12 → | |
| Qwen3.5-4B | Qwen | — | — | 262.1K | — | — | active | 27 Feb 2026 | reconstructed | as of 2026-03-12 → |
| Qwen3.5-9B | Qwen | — | — | 262.1K | — | — | active | 27 Feb 2026 | reconstructed | as of 2026-03-12 → |
| Nano Banana 2 (Gemini 3.1 Flash Image Preview) | — | — | 65.5K | — | — | active | 26 Feb 2026 | reconstructed | as of 2026-03-12 → | |
| Seed-2.0-Mini | ByteDance | — | — | 262.1K | — | — | active | 26 Feb 2026 | reconstructed | as of 2026-03-12 → |
| Gemini 3.1 Pro Preview Custom Tools | — | — | 1.05M | — | — | active | 25 Feb 2026 | reconstructed | as of 2026-03-12 → | |
| Qwen3.5-122B-A10B | Qwen | — | — | 262.1K | — | — | active | 25 Feb 2026 | reconstructed | as of 2026-03-12 → |
| Qwen3.5-27B | Qwen | — | — | 262.1K | — | — | deprecated | 25 Feb 2026 | reconstructed | as of 2026-03-12 → |
| Qwen3.5-35B-A3B | Qwen | — | — | 262.1K | — | — | deprecated | 25 Feb 2026 | reconstructed | as of 2026-03-12 → |
| Qwen3.5-Flash | Qwen | — | — | 1M | — | — | active | 25 Feb 2026 | reconstructed | as of 2026-03-12 → |
| Step-3.5-Flash-REAP-121B-A11B | Cerebras Systems | — | — | — | — | — | active | 25 Feb 2026 | reconstructed | as of 2026-03-12 → |
| Step-3.5-Flash-REAP-149B-A11B | Cerebras Systems | — | — | — | — | — | active | 25 Feb 2026 | reconstructed | as of 2026-03-12 → |
| gpt-5.3-codex | OpenAI | — | — | 400K | — | — | active | 24 Feb 2026 | reconstructed | as of 2026-03-12 → |
| pythia-14m | EleutherAI | — | — | — | — | — | active | 24 Feb 2026 | reconstructed | as of 2026-03-12 → |
Showing 25 of 711 models that existed on this date (API cap 25). Open a model's History tab for its full as-of record.
“Today” columns compare the as-of value with the current record; “same” means no recorded change. Click any as-of value for the claim behind it (source, tier, valid from → to).
Provider prices
Offers valid on 12 Mar 2026 0
| Model | Provider | Input / 1M | Output / 1M | Context | Valid from | Valid to | Source |
|---|---|---|---|---|---|---|---|
| No price row's validity interval covers this date (price history starts when AI Atlas first observed the provider). | |||||||
offers whose validity interval covers the date (price rows are append-only)
Benchmark leaders
Leaders known by 12 Mar 2026 0
| Benchmark | Leader | Organization | Score | Metric · group | Trust | Models ranked |
|---|---|---|---|---|---|---|
| No benchmark result had been observed by this date — leaders are known only once a result is observed (evaluation dates are not used). | ||||||
One row per benchmark: the best current row of the primary comparability group among results observed by the date. See the benchmark's Frontier tab for the full leader history.
leaders from results observed by the date (evaluation dates are not used: a result is known only once observed)
Hardware
Hardware as of 12 Mar 2026 25
| Hardware | Manufacturer | Kind | Memory | Bandwidth | Released | Basis |
|---|---|---|---|---|---|---|
| Apple M3 Ultra | Apple | soc | 96 / 256 / 512 GB | 819 GB/s | Mar 2025 | reconstructed |
| NVIDIA GeForce RTX 5090 | NVIDIA | gpu | 32 GB | 1,792 GB/s | Jan 2025 | reconstructed |
| NVIDIA DGX Spark | NVIDIA | system | 128 GB | 273 GB/s | 2025 | reconstructed |
| AMD Instinct MI325X | AMD | gpu | 256 GB | 6,000 GB/s | Oct 2024 | reconstructed |
| Apple M4 Max | Apple | soc | 36 / 48 / 64 / 128 GB | 546 GB/s | Oct 2024 | reconstructed |
| Apple M4 Pro | Apple | soc | 24 / 48 / 64 GB | 273 GB/s | Oct 2024 | reconstructed |
| Apple M4 | Apple | soc | 16 / 24 / 32 GB | 120 GB/s | May 2024 | reconstructed |
| NVIDIA B200 | NVIDIA | gpu | 180 GB | — | Mar 2024 | reconstructed |
| NVIDIA H200 | NVIDIA | gpu | 141 GB | 4,800 GB/s | 2024 | reconstructed |
| AMD Instinct MI300X | AMD | gpu | 192 GB | 5,300 GB/s | Dec 2023 | reconstructed |
| Apple M3 | Apple | soc | 8 / 16 / 24 GB | 100 GB/s | Oct 2023 | reconstructed |
| Apple M3 Pro | Apple | soc | 18 / 36 GB | 150 GB/s | Oct 2023 | reconstructed |
| Apple M3 Max | Apple | soc | 36 / 48 / 64 / 96 / 128 GB | 400 GB/s | Oct 2023 | reconstructed |
| Apple M2 Ultra | Apple | soc | 64 / 128 / 192 GB | 800 GB/s | Jun 2023 | reconstructed |
| Apple M2 Max | Apple | soc | 32 / 64 / 96 GB | 400 GB/s | Jan 2023 | reconstructed |
| Apple M2 Pro | Apple | soc | 16 / 32 GB | 200 GB/s | Jan 2023 | reconstructed |
| NVIDIA GeForce RTX 4090 | NVIDIA | gpu | 24 GB | 1,008 GB/s | Oct 2022 | reconstructed |
| NVIDIA H100 SXM | NVIDIA | gpu | 80 GB | 3,350 GB/s | Sept 2022 | reconstructed |
| Apple M2 | Apple | soc | 8 / 16 / 24 GB | 100 GB/s | Jun 2022 | reconstructed |
| Apple M1 Ultra | Apple | soc | 64 / 128 GB | 800 GB/s | Mar 2022 | reconstructed |
| Apple M1 Pro | Apple | soc | 16 / 32 GB | 200 GB/s | Oct 2021 | reconstructed |
| Apple M1 Max | Apple | soc | 32 / 64 GB | 400 GB/s | Oct 2021 | reconstructed |
| Apple M1 | Apple | soc | 8 / 16 GB | 68 GB/s | Nov 2020 | reconstructed |
| NVIDIA A100 80GB | NVIDIA | gpu | 80 GB | 2,039 GB/s | Nov 2020 | reconstructed |
| NVIDIA GeForce RTX 3090 | NVIDIA | gpu | 24 GB | 936 GB/s | Sept 2020 | reconstructed |
Shareable: this URL reproduces the view. Per-entity: every entity page has a History tab with the same as-of reconstruction. What changed since? Diff 12 Mar 2026 → today. Methodology →