Independent, verifiable rankings of GPT, Claude, Gemini, Llama, DeepSeek and 100+ frontier AI models. Evaluated continuously by Composite TrueSkill, coding benchmarks, output speed, and live API token pricing.
Code generation, debugging & refactoring
Prose, tone matching & creative content
Analysis, synthesis & deep logical tasks
Text-to-image diffusion & editing
Blended token cost (8:1 input/output ratio) plotted against TrueSkill score. Models on the green line are Pareto-efficient.
| Rank | Model & Provider | Score | Reasoning | Coding | Arena ELO | Params | Context | Speed | Pricing $/1M | Action |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | GPT-5.6 Sol OpenAI | 57.4 | 56.8 | 50.6 | 2,134 | MoE (~1.8T) | 1.1M | 102 c/s | $7.78 | Compare |
| 2 | Claude Opus 5 Anthropic | 56.2 | 55.3 | 42.7 | 2,668 | Dense (~220B) | 1.0M | 58 c/s | $7.22 | Compare |
| 3 | Claude Fable 5 Anthropic | 56.1 | 53.5 | 48.8 | 2,003 | MoE (~350B) | 1.0M | 78 c/s | $14.40 | Compare |
| 4NEW | Claude Mythos Preview Anthropic | 56 | 56.8 | 46.6 | 2,190 | Research Tier | 1.0M | 45 c/s | Free | Compare |
| 5 | Kimi K3 Moonshot AI | 54.9 | 53.9 | 45.9 | 1,816 | 2.8T (MoE) | 1.0M | 135 c/s | $4.33 | Compare |
| 6NEW | GLM-5.3 Zhipu AI | 54.7 | 54.9 | 45.4 | 1,780 | 753B (MoE) | 1.0M | 142 c/s | $1.73 | Compare |
| 7NEW | DeepSeek-V4-Pro-0813 DeepSeek | 54.5 | 52 | 44.3 | 1,980 | 1.6T (MoE) | 1.0M | 199 c/s | $0.48 | Compare |
| 8NEW | Qwen3.8 Max Alibaba Cloud / Qwen Team | 53.2 | 52 | 42.1 | 1,850 | 2.4T (MoE) | 1.0M | 160 c/s | $0.85 | Compare |
| 9 | GPT-5.6 Terra OpenAI | 52.8 | 51.1 | 46.4 | 1,185 | MoE (~500B) | 1.1M | 119 c/s | $3.11 | Compare |
| 10 | Claude Opus 4.8 Anthropic | 51.9 | 51.4 | 43.8 | 1,689 | Dense | 1.0M | 125 c/s | $7.22 | Compare |
| 11 | Muse Spark 1.1 Meta | 51.7 | 52.3 | 37.8 | 1,070 | Dense (~70B) | 1.0M | 229 c/s | $1.58 | Compare |
| 12NEW | Gemini 3.7 Flash Google | 51.1 | 49.9 | 38.6 | 1,720 | Sparse MoE | 1.0M | 621 c/s | $1.08 | Compare |
| 13 | Claude Sonnet 5 Anthropic | 49.6 | 49 | 40 | 1,588 | Dense (~100B) | 1.0M | 59 c/s | $2.89 | Compare |
| 14 | GPT-5.5 OpenAI | 49.3 | 48.6 | 41.1 | 2,124 | MoE | 1.1M | 49 c/s | $7.78 | Compare |
| 15NEW | DeepSeek-V4-Flash-Vision-Exp DeepSeek | 48.7 | 45.3 | 38.5 | 1,650 | Vision MoE | 1.0M | 310 c/s | $0.27 | Compare |
| 16NEW | Grok 4.5 xAI | 48.2 | 47.8 | 37.2 | 1,540 | MoE (~300B) | 2.0M | 110 c/s | $2.00 | Compare |
Find the optimal model for specific tasks, modalities, or industry verticals.
DeepSeek's first official V4-Flash vision API. Same $0.22/$0.66 off-peak list as V4-Flash, 1M context. Self-reported Chartography 64.3 and TerminalBench 83.9.
Gemini 3.7 Flash scores 65.3% on DeepSWE v1.1 and 43.6% on FrontierCode 1.1 Main at $0.35/$1.50 per million tokens with record 621 tok/s.
DeepSeek-V4-Pro-0813 is the GA version behind API id deepseek-v4-pro. $0.435 / $0.003625 cached with 1.6T MoE architecture.
Grok 4.6 scores 1753 on GDPVal-AA v2 and 65.9% on DeepSWE v1.1, with 2.0M context and $2.00 blended pricing.