Head-to-head · live
GPT vs Claude
Side-by-side comparison of the current flagship models from OpenAI and Anthropic across every benchmark that matters — refreshed live, no marketing spin.
OpenAI
OpenAI: GPT-6 Astra ProAnthropic
Anthropic: Claude Fable 5.1Row-by-row
| Benchmark | OpenAI: GPT-6 Astra Pro | Anthropic: Claude Fable 5.1 | Winner |
|---|---|---|---|
| Arena Elo | 1632 | 1642 | Claude |
| AA Index | 78.0 | 79.0 | Claude |
| LiveBench | 87.2 | 86.9 | GPT |
| GPQA | 92.1% | 91.2% | GPT |
| SWE-bench | 81.7% | 82.4% | Claude |
| Terminal-Bench | 71.3% | 71.8% | Claude |
| Aider Polyglot | 87.2% | 88.1% | Claude |
| Output $/1M | $50.00 | $50.00 | tie |
Frequently asked
- GPT vs Claude — which is better?
- It depends on the task. Claude flagships (Opus/Fable/Mythos) consistently lead on SWE-bench and long-form writing quality; GPT flagships (Sol/Terra) lead on raw reasoning benchmarks and multimodality. The tables above show the current numbers head-to-head.
- Is Claude smarter than GPT?
- On some benchmarks yes, on others no. Recent Claude flagships lead on SWE-bench Verified (coding) and Terminal-Bench (agentic), while OpenAI's Sol line leads on GPQA and ARC-AGI. See the row-by-row comparison above.
- Which is cheaper, GPT or Claude?
- GPT-family flagships and Claude flagships are priced within ~30% of each other for output tokens. Smaller siblings (mini, haiku, flash) are 10× cheaper on both sides. See /cheapest for the full price ranking.
Also see: best LLM, best coding LLM, top-5 comparison.