Benchmarks · 2026-10-15
Do the best-scoring models win?
Benchmark results from OpenRouter's unified endpoint: Artificial Analysis composite indexes, Design Arena head-to-head Elo ratings, and OpenRouter's own evals with cost and latency per task. Each is cross-referenced with the token share those models receive on OpenRouter today.
Top 20 models — Artificial Analysis · Intelligence
Higher is better (Intelligence). Bars coloured by lab.
Efficiency frontier — Intelligence vs output price
Bubble = usage share. The line traces the Pareto frontier: models where nothing cheaper scores as high.
Benchmark score vs OpenRouter share — Intelligence
Top-right = high score and heavily used · bottom-right = high score but under-adopted · top-left = popular despite a lower score. Models not in the top 50 by usage sit on the floor.
Best for web search agents
BrowseComp: accuracy, average cost and latency per task, with each model's OpenRouter token share.
Claude Opus 5 leads BrowseComp at 89.2%; GPT-5.6-Sol gets within 10% of that for $0.503 per task.
| Model | Search engine | Accuracy | $ / task | Latency / task | Token share |
|---|---|---|---|---|---|
| Claude Opus 5 | parallel | 89.2% | $2.36 | 2.3 min | 0.39% |
| GPT-5.6-Sol | perplexity | 82.4% | $0.503 | 1.9 min | 0.40% |
| Deepseek V4 Flash 0731 | perplexity | 77.0% | $0.076 | 2.3 min | — |
| GPT-5.6-Luna | perplexity | 74.0% | $0.099 | 1.9 min | 1.07% |
Web search agents
Claude Opus 5 tops OpenRouter's BrowseComp search eval at 89.2% accuracy, $2.36 per task. Source: OpenRouter evals (openrouter.ai) via OpenRouter (openrouter.ai/rankings), as of 2026-10-15. tidelines.ai/benchmarks
What a session actually costs in Hermes Agent
Median cost of one coding session by session length. 30-day window, log scale.
GPT-6 Astra Pro costs 489× more per session than Gemini 2.5 Flash Lite in this agent.
How to read: Each row is a model. Dots run from short sessions to long ones — the further right, the more a median session costs. The number on the right is the 10–49 turns median.
- $8.49
- $3.81
- $3.23
- $2.93
- $2.92
- $2.06
- $1.53
- $1.32
- $1.30
- $0.994
- $0.992
- $0.869
Intelligence per dollar of real work
Benchmark score against the median 10–49 turns session cost in Hermes Agent.
Models on the dashed line are undominated: nothing is both cheaper and better-scoring. Bubble size is OpenRouter token share.
How to read: Cost runs left to right on a log scale, benchmark score bottom to top. Up and to the left is better value. Click a bubble for the model page.