Model leaderboard
3qk7's own evidence index, recomputed from direct LiveBench, Artificial Analysis and Agent Arena metrics. Price and API availability are context only.
Overall
A relative evidence index combining long-horizon tasks, independent intelligence measurements and real agent-work signals.
| Rank | Model | Evidence index | Coverage | API |
|---|---|---|---|---|
| 01 | Claude Opus 5.5Anthropic | 95.0 | 100% | Available$4 / $20 / 1M |
| 02 | Claude Fable 5.1Anthropic | 83.1 | 100% | Available$10 / $50 / 1M |
| 03 | GPT-6 AstraOpenAI | 75.1 | 100% | Available$10 / $50 / 1M |
| 04 | Claude Sonnet 5.5Anthropic | 57.7 | 100% | Available$2 / $10 / 1M |
| 05 | Gemini 4 ArgonGoogle | 57.1 | 100% | Available$2 / $10 / 1M |
| 06 | GPT-6.1 SolOpenAI | 55.4 | 100% | Available$2 / $10 / 1M |
| 07 | Claude Fable 5Anthropic | 46.0 | 70% | Available$10 / $50 / 1M |
| 08 | Claude Opus 5Anthropic | 34.0 | 70% | Available$5 / $25 / 1M |
| 09 | GPT-6 SolOpenAI | 33.0 | 70% | Available$2 / $10 / 1M |
| 10 | DeepSeek V4.1 Flash 1017DeepSeek | 7.0 | 100% | Available$0.15 / $0.6 / 1M |
| 11 | Muse WonderMeta | 6.4 | 100% | Available$4 / $20 / 1M |
How to read this board
Each raw metric is converted to a midrank percentile within the current model cohort, then combined by board weights. Missing evidence gets a neutral 50th-percentile placeholder but lowers coverage; below 70% weighted coverage a model stays unranked.
Overall: LiveBench 50% · Artificial Analysis 30% · Agent Arena Work 20%. Category boards use their own published weights.
No aggregate score or rank from another leaderboard is imported. NewLab freezes source values and the formula; 3qk7 recomputes this snapshot from those direct metrics.
Method version: 3qk7-evidence-index-v0.1