Database section · 59 records shown
AI Models
Language, reasoning, coding, image, video, speech, robotics and science models — with release dates, disclosed parameters, context windows and weights status. Undisclosed means undisclosed.
Field notes
The dates column tells this section's story better than any essay: the gap between frontier releases has collapsed from years to weeks. Open the 2026 records first and note who discloses what — secrecy itself became a spec.
Start with these three
C5
Claude 5 (Fable 5 / Opus 5 / Sonnet 5)
The 2026 frontier line — four releases in under two months.
DV
DeepSeek-V4
1.6 trillion parameters, MIT-licensed. The efficiency school, scaled up.
MM
Meta Muse Spark
The record that captures the end of the Llama era — and a strategy reversal.
Key columns: parameters are the model's learned weights (open models disclose them, closed ones usually don't); context is how much text the model can consider at once; open weights can be downloaded, closed models are API-only.
| Model | Type | Released | Parameters | Context | Weights | ★ |
|---|---|---|---|---|---|---|
| Meta Muse SparkMeta (Superintelligence Labs) | Multimodal | 2026-08 | Not publicly disclosed | 1M (Spark 1.2) | Closed (Glimmer open; Spark 1.2 opening pledged) | |
| Gemini 3.5–3.7 Flash (2026 line)Google DeepMind | Multimodal | 2026-08 | Not publicly disclosed | 1M | Closed | |
| Qwen 3.6 / 3.7-Max / 3.8 (2026 line)Alibaba | Language | 2026-08 | up to 2.4T (95B active) | 1M | Open (Apache) + closed Max tier | |
| Claude 5 (Fable 5 / Opus 5 / Sonnet 5)Anthropic | Coding | 2026-07 | Not publicly disclosed | 1M | Closed | |
| GPT-5.6 (and the 2026 GPT-5.x line)OpenAI | Language | 2026-07 | Not publicly disclosed | 400K | Closed | |
| DeepSeek-V4DeepSeek | Language | 2026-04 | 1.6T (49B active) | 1M | Open (MIT) | |
| Gemini 3Google DeepMind | Multimodal | 2025-11 | Not publicly disclosed | 1M | Closed | |
| Claude 4.5 (Opus / Sonnet / Haiku)Anthropic | Coding | 2025-11 | Not publicly disclosed | 200K–1M | Closed | |
| Marble (World Labs)World Labs | World Model | 2025-11 | Not publicly disclosed | — | Closed | |
| GLM-4.6Zhipu AI (Z.ai) | Coding | 2025-10 | 355B (32B active) | 200K | Open | |
| GPT-5OpenAI | Language | 2025-08 | Not publicly disclosed | 400K | Closed | |
| Genie 3Google DeepMind | World Model | 2025-08 | Not publicly disclosed | — | Closed | |
| gpt-ossOpenAI | Language | 2025-08 | 117B (5.1B active) / 21B | 128K | Open | |
| Grok 4xAI | Reasoning | 2025-07 | Not publicly disclosed | 256K | Closed | |
| Kimi K2Moonshot AI | Language | 2025-07 | 1T (32B active) | 256K | Open | |
| MiniMax-M1MiniMax | Reasoning | 2025-06 | 456B (46B active) | 1M | Open | |
| Mistral Large & MagistralMistral AI | Reasoning | 2025-06 | 24B (open) / larger closed | 128K | Mixed | |
| Claude 4 (Opus 4 / Sonnet 4)Anthropic | Coding | 2025-05 | Not publicly disclosed | 200K–1M | Closed | |
| Veo 3Google DeepMind | Video | 2025-05 | Not publicly disclosed | — | Closed | |
| Qwen3Alibaba | Language | 2025-04 | 0.6B–235B (22B active) | 128K–1M | Open | |
| Llama 4Meta | Multimodal | 2025-04 | 109B–400B (17B active) | up to 10M (claimed) | Open | |
| Midjourney v7Midjourney | Image | 2025-04 | Not publicly disclosed | — | Closed | |
| OpenAI o3OpenAI | Reasoning | 2025-04 | Not publicly disclosed | 200K | Closed | |
| Cohere Command ACohere | Language | 2025-03 | 111B | 256K | Open (research) / commercial license | |
| Gemini 2.5Google DeepMind | Reasoning | 2025-03 | Not publicly disclosed | 1M | Closed | |
| Gemma 3Google DeepMind | Language | 2025-03 | 1B–27B | 128K | Open | |
| NVIDIA Isaac GR00T N1NVIDIA | Robotics | 2025-03 | 2B | — | Open | |
| NVIDIA NemotronNVIDIA | Language | 2025-03 | 4B–253B | 128K | Open | |
| Runway Gen-4Runway | Video | 2025-03 | Not publicly disclosed | — | Closed | |
| Figure HelixFigure AI | Robotics | 2025-02 | 7B + 80M | — | Closed | |
| Grok 3xAI | Language | 2025-02 | Not publicly disclosed | 1M | Closed | |
| Wan 2.xAlibaba | Video | 2025-02 | 1.3B–27B | — | Open | |
| DeepSeek-R1DeepSeek | Reasoning | 2025-01 | 671B (37B active) | 128K | Open | |
| NVIDIA CosmosNVIDIA | World Model | 2025-01 | 4B–14B | — | Open | |
| DeepSeek-V3DeepSeek | Language | 2024-12 | 671B (37B active) | 128K | Open | |
| Amazon NovaAmazon | Multimodal | 2024-12 | Not publicly disclosed | 300K | Closed | |
| Microsoft Phi-4Microsoft | Language | 2024-12 | 14B | 16K | Open | |
| SoraOpenAI | Video | 2024-12 | Not publicly disclosed | — | Closed | |
| π0 (pi-zero)Physical Intelligence | Robotics | 2024-10 | ~3B | — | Open | |
| OpenAI o1OpenAI | Reasoning | 2024-09 | Not publicly disclosed | 128K | Closed | |
| FLUX.1Black Forest Labs | Image | 2024-08 | 12B | — | Open (dev/schnell) + closed pro | |
| Llama 3.1 405BMeta | Language | 2024-07 | 405B dense | 128K | Open | |
| Claude 3.5 SonnetAnthropic | Coding | 2024-06 | Not publicly disclosed | 200K | Closed | |
| ESM3EvolutionaryScale | Science | 2024-06 | 98B | — | Open (small) / API (large) | |
| KlingKuaishou | Video | 2024-06 | Not publicly disclosed | — | Closed | |
| AlphaFold 3Google DeepMind | Science | 2024-05 | — | — | Open (AF2) / server + code w/ restrictions (AF3) | |
| GPT-4oOpenAI | Multimodal | 2024-05 | Not publicly disclosed | 128K | Closed | |
| AI21 JambaAI21 Labs | Language | 2024-03 | 52B–398B MoE hybrids | 256K | Open | |
| Claude 3 (Opus / Sonnet / Haiku)Anthropic | Language | 2024-03 | Not publicly disclosed | 200K | Closed | |
| Gemini 1.5Google DeepMind | Multimodal | 2024-02 | Not publicly disclosed | 1M–2M | Closed | |
| OpenAI text-embedding-3OpenAI | Embedding | 2024-01 | Not publicly disclosed | — | Closed | |
| Mistral 7B & MixtralMistral AI | Language | 2023-09 | 7B / 47B (13B active) | 32K | Open | |
| Llama 2Meta | Language | 2023-07 | 7B–70B | 4K | Open | |
| GPT-4OpenAI | Language | 2023-03 | Not publicly disclosed | 8K–128K | Closed | |
| WhisperOpenAI | Speech & Music | 2022-09 | 39M–1.5B | — | Open | |
| Stable DiffusionStability AI / CompVis / Runway | Image | 2022-08 | 0.9B–8B | — | Open | |
| GPT-3OpenAI | Historical | 2020-06 | 175B | 2K | Closed | |
| GPT-2OpenAI | Historical | 2019-02 | 1.5B | 1K | Open | |
| AlphaGo / AlphaZeroGoogle DeepMind | Historical | 2016-03 | — | — | Closed |
"Not publicly disclosed" is shown on record pages where a developer has not published a figure — parameters for closed frontier models are usually undisclosed. Download: models.csv · JSON