Amazon Nova (Frontier Model Research)
2 tracked models, none with public benchmark components yet — no Club Score until a sourced score exists. Latest: Nova 2 Pro (Dec 2, 2025).
AI Labs
The Labs Leaderboard.
The organisations behind the models, ranked like clubs: every number below is a live aggregate of the model database — best model, squad depth, cheapest capable model, open-weight share.
18Labs profiled
14Clubs ranked
AnthropicLeague-leading club (95.8)
15%Open-Weight Index — share of the League top 20 that is open-weight (▼ 5.0)
| # | Lab | Club Score | Best model | Top-3 avg | Squad | Cheapest capable | Latest release | Open % | Form |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Anthropic Closed | 95.8 | Claude Fable 5.1 97 | 93.5 | 4 | Claude Haiku 5.5 $0.20/1M blended | Claude Haiku 5.5 Oct 7, 2026 | 0% | — 0.0 |
| 2 | OpenAI Closed | 94.4 | GPT-6 Astra 97 | 90.9 | 6 | GPT-5.6 Luna $0.45/1M blended | GPT-6.1 Sol Sep 29, 2026 | 0% | — 0.0 |
| 3 | Google DeepMind Mixed | 84.2 | Gemini 4 Argon 91 | 74.5 | 5 | Gemini 3.5 Flash-Lite $0.85/1M blended | Gemini 4 Argon Oct 13, 2026 | 40% | ▼ 1.4 |
| 4 | DeepSeek Fully open | 73.0 | DeepSeek V4.1 Flash 75 | 69.3 | 2 | DeepSeek V4.1 Flash $0.52/1M blended | DeepSeek V4.1 Flash Sep 10, 2026 | 100% | ▲ 2.6 |
| 5 | xAI (SpaceXAI) Closed | 72.3 | Grok 4.3 74 | 70.3 | 2 | Grok 4.3 $1.56/1M blended | Grok 4.7 Sep 21, 2026 | 0% | — 0.0 |
| 6 | Meta Superintelligence Labs Mixed | 71.6 | Muse Wonder 76 | 64.3 | 5 | Muse Spark 1.3 $2.00/1M blended | Muse Wonder Oct 12, 2026 | 60% | ▲ 16.6 |
| 7 | Moonshot AI Fully open | 68.3 | Kimi K2.6 70 | 66.2 | 3 | Kimi K2.6 $1.71/1M blended | Kimi K3 Jul 16, 2026 | 100% | ▲ 7.3 |
| 8 | Alibaba Qwen (Tongyi Lab) Fully open | 61.5 | Qwen3.8-Max 62 | 61.2 | 2 | Qwen3.8-Max $3.00/1M blended | Qwen3.8-Max Aug 3, 2026 | 100% | ▲ 3.5 |
| 9 | Mistral AI Fully open | 59.1 | Mistral Medium 3.5 59 | 59.1 | 2 | Mistral Small 4 $0.26/1M blended † | Mistral Medium 3.5 Apr 28, 2026 | 100% | ▲ 28.8 |
| 10 | NVIDIA Fully open | 45.7 | Nemotron 3 Ultra 550B-A55B 57 | 28.6 | 3 | — | Nemotron 3 Ultra 550B-A55B Jun 4, 2026 | 100% | ▲ 13.5 |
| 11 | Thinking Machines Lab Fully open | 43.1 | Inkling 43 | 43.1 | 2 | — | Inkling Small Jul 31, 2026 | 100% | — |
| 12 | Microsoft AI (MAI) Closed | 1.3 | MAI-Code-1-Flash 1 | 1.3 | 2 | — | MAI-Thinking-1 Jun 2, 2026 | 0% | — |
| 13 | Cohere Fully open | 0.0 | Command A+ 0 | 0.0 | 2 | Command A $4.38/1M blended † | Command A+ May 20, 2026 | 100% | — |
| 14 | AI21 Labs API sunset Fully open | 0.0 | Jamba Large 1.7 0 | 0.0 | 2 | — | Jamba Large 1.7 Jan 15, 2026 | 100% | — |
Club Score = 0.6 × the lab’s best model’s League Score + 0.4 × the mean of its top three. † = no model from this lab clears the League Score 60 “capable” bar with pricing, so its cheapest priced model is shown. Form compares Club Score with the previous weekly snapshot. Full scoring rules: methodology.
Tracked, not yet scored
Research-only labs
Real labs, honestly unranked: no models in the League database, so no Club Score is invented for them.
Perplexity
Its Sonar models are search-tuned derivatives that are not benchmark-tracked like frontier models — profile only for now.
Lab profile →Stability AI
Its models are media models (image, audio, video), which sit outside this database's LLM pricing and benchmark schema.
Lab profile →Safe Superintelligence Inc. In stealth
Zero public models, by design — an honest research-only profile.
Lab profile →Labs by country
USA
Amazon, Anthropic, Meta AI, Microsoft AI, NVIDIA, OpenAI, Perplexity, SSI, Thinking Machines, xAI
10 labs · best Club Score 95.8 · cheapest capable: Claude Haiku 5.5 ($0.20/1M blended)
China
3 labs · best Club Score 73.0 · cheapest capable: DeepSeek V4.1 Flash ($0.52/1M blended)
United Kingdom
2 labs · best Club Score 84.2 · cheapest capable: Gemini 3.5 Flash-Lite ($0.85/1M blended)
France
1 lab · best Club Score 59.1 · cheapest capable: Mistral Small 4 ($0.26/1M blended)
Canada
1 lab · best Club Score 0.0 · cheapest capable: Command A ($4.38/1M blended)
Israel
1 lab · best Club Score 0.0