Every LLM benchmark, one honest table.
Scores, prices, and speed for 185 models — aggregated from public sources, with the origin and provenance of every number visible. Refreshed automatically, last 5h ago.
Frontier right now
Full leaderboard →| # | Model | Lab | Context | $/1M in · out | Intelligence | Coding | Agentic | Arena Elo |
|---|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Anthropic | 1M | $8 · $40 | 64.6 | — | — | — |
| 2 | GPT-6.2 Astra | OpenAI | 1.1M | $8 · $40 | 62.1 | — | — | — |
| 3 | Gemini 4 Neon | 1.0M | $3 · $15 | 61.4 | — | — | — | |
| 4 | Gemini 4 Carbon | 1.0M | $2 · $10 | 60.0 | — | — | — | |
| 5 | Claude Opus 5.5 | Anthropic | 1M | $4 · $20 | 59.1 | — | — | 1516 |
| 6 | GPT-6.2 Sol | OpenAI | 1.1M | $2 · $10 | 57.3 | — | — | — |
| 7 | GPT-6 Astra | OpenAI | 1.1M | $10 · $50 | 56.9 | 76.9 | 51.0 | — |
| 8 | GPT-6.1 Sol | OpenAI | 1.1M | $2 · $10 | 54.6 | — | — | — |
| 9 | Claude Sonnet 5.5 | Anthropic | 1M | $2 · $10 | 54.5 | — | — | 1470 |
| 10 | Gemini 4 Hydrogen | 1.0M | $1.5 · $7.5 | 53.9 | — | — | — | |
| 11 | GPT-6.1 Terra | OpenAI | 1.1M | $1 · $5 | 51.3 | — | — | — |
| 12 | Gemini 4 Argon | 1.0M | $2 · $10 | 50.4 | — | — | 1533 | |
| 13 | Claude Fable 5.1 | Anthropic | 1M | $10 · $50 | 50.0 | 81.6 | 57.9 | — |
| 14 | Claude Opus 5 | Anthropic | 1M | $5 · $25 | 49.7 | 78.0 | 56.5 | 1502 |
| 15 | Claude Fable 5 | Anthropic | 1M | $10 · $50 | 49.6 | 76.5 | 50.7 | 1493 |
Intelligence, coding & agentic indices by Artificial Analysis; Arena Elo by LMArena (CC BY 4.0). Best value across model variants shown.
Intelligence vs. price
The frontier you actually pay for — up and to the left is better. Hover a point for details; click it to open the model. Free-tier models excluded.
API-onlyOpen weights
New models
- Gemini 4 NeonGoogleNov 18, 2026
- GPT-6.2 SolOpenAINov 17, 2026
- GPT-6.2 AstraOpenAINov 17, 2026
- Gemini 4 HydrogenGoogleNov 12, 2026
- Gemini 4 CarbonGoogleNov 10, 2026
- Muse EmberMetaNov 9, 2026
- Claude Fable 5.2AnthropicNov 9, 2026
- GPT-6.1 TerraOpenAIOct 26, 2026
How to read the numbers
- independent — measured by an independent evaluator
- crowd — human preference votes (Elo)
- mirror — mirrored via an aggregator API
- vendor — self-reported by the model's vendor
Every score keeps a link to where it came from and when we saw it. Read the methodology.