Benchmarks and price
Independent scores next to each model's official API price. A higher score does not make a model right for your task: use these to shortlist, then test on your own prompts. Scores from LMArena and Epoch AI, CC BY 4.0; see sources.
| Model | Arena text | Arena coding | FrontierMath | GPQA Diamond | SimpleQA Verified | Price / 1M |
|---|---|---|---|---|---|---|
| Gemini 4 Argon best value Google · Most capable | 1529 ±6High | 1538 ±12High | 86.0%high | 96.4%high | 76.0%high | $4.00 |
| Claude Opus 5.5 Anthropic · Most capable | 1516 ±7High | 1547 ±12High | 91.2%max | 90.6%max | 72.2%max | $8.00 |
| Claude Fable 5.1 Anthropic · Frontier | 1514 ±6Max | 1530 ±11Max | 90.2%max | — | 70.8%max | $20.00 |
| Claude Opus 5 Anthropic | 1508 ±5Max | 1533 ±6High | 85.6%max | 93.9%max | 59.9%max | $10.00 |
| Claude Opus 4.6 Anthropic | 1502 ±3High | 1533 ±5 | 66.0%max | 90.5%32K | 47.0%max | $10.00 |
| Gemini 3.8 Flash best value Google · Balanced | 1498 ±4High | 1509 ±7High | 68.4%high | 95.4%high | 69.7%high | $1.50 |
| Claude Fable 5 Anthropic | 1492 ±4High | 1518 ±6High | 87.0%max | 85.9%max | 70.7%xhigh | $20.00 |
| Claude Opus 4.7 Anthropic | 1490 ±4High | 1519 ±6High | 70.2%max | 90.2%xhigh | 51.7%xhigh | $10.00 |
| gemini-3.1-pro-preview Google | 1480 ±3 | 1484 ±4 | 59.6% | 94.4%high | 73.5%high | $4.50 |
| Gemini 3.6 Flash Google | 1480 ±4High | 1491 ±7High | 58.9%high | 94.1%high | 66.2%high | $1.50 |
| Kimi K3 Moonshot AI | 1476 ±4Max | 1508 ±7Max | 72.2%max | 93.1%max | 50.6%max | $6.00 |
| Z.ai GLM 5.3 Z.ai (Zhipu) · Most capable | 1474 ±5Max | 1497 ±8Max | 68.8%max | 90.9%max | 41.0%max | $2.15 |
| Z.ai GLM 5.3 Flash best value Z.ai (Zhipu) · Fast & low-cost | 1473 ±5 | 1509 ±7 | 55.8%max | 90.2%max | — | $0.2375 |
| Claude Sonnet 5.5 Anthropic · Balanced | 1473 ±7xHigh | 1512 ±13xHigh | 88.8%max | 95.6%max | 46.5%max | $4.00 |
| gpt-5.5 OpenAI | 1472 ±4High | 1494 ±6High | 85.3%xhigh | 90.7%low | 63.0%xhigh | $11.25 |
| GLM-5.2 Z.ai (Zhipu) | 1470 ±4Max | 1487 ±6Max | 59.2%max | 91.9%max | 34.2%max | $2.15 |
| gpt-5.4 OpenAI | 1469 ±4High | 1496 ±6High | — | — | — | $5.625 |
| Claude Opus 4.8 Anthropic | 1461 ±4High | 1491 ±6High | 80.0%max | 91.0%max | 53.0%max | $10.00 |
| Claude Sonnet 4.6 Anthropic | 1458 ±3 | 1503 ±6 | — | 87.4%32K | 35.5%high | $6.00 |
| Gemini 2.5 Pro Google | 1458 ±2 | 1451 ±4 | 24.6% | 85.3% | — | $3.4375 |
| gpt-5.6-sol OpenAI | 1457 ±4xHigh | 1495 ±7xHigh | 89.1%max | 93.5%max | 69.7%max | $8.00 |
| Qwen3.7 Plus Qwen (Alibaba) · Balanced | 1455 ±4 | 1473 ±6 | 34.4%none | 87.9% | — | $0.56 |
| Claude Opus 4.5 Anthropic | 1451 ±320251101 | 1503 ±7High, 32K, 20251101 | — | — | — | $10.00 |
| gpt-5.6-terra OpenAI | 1448 ±4xHigh | 1485 ±6xHigh | 86.0%max | 93.3%max | 43.2%max | $4.50 |
| grok-4.5 xAI | 1448 ±4 | 1474 ±7 | 57.2%high | 93.4%high | 48.3%high | $3.00 |
| gpt-6.1-sol OpenAI · Most capable | 1446 ±7Max | 1486 ±14Max | 93.7%max | 95.4%max | 73.9%max | $4.00 |
| gpt-5.1 OpenAI | 1443 ±4High | 1453 ±7High | — | — | — | $3.4375 |
| Claude Sonnet 5 Anthropic | 1443 ±4High | 1484 ±6High | 65.6%max | 90.5%xhigh | 33.7%max | $4.00 |
| gpt-6-astra OpenAI · Frontier | 1441 ±6Max | 1488 ±11Max | 93.7%max | 95.8%max | 75.6%max | $20.00 |
| Claude Sonnet 4.5 Anthropic | 1439 ±320250929 | 1488 ±5High, 32K, 20250929 | — | — | — | $6.00 |
| Qwen/Qwen3.8-27B Qwen (Alibaba) | 1438 ±5 | 1481 ±8 | — | — | — | $1.60 |
| Qwen3.6 Plus Qwen (Alibaba) | 1436 ±4 | 1467 ±6 | 38.2% | 88.4% | 44.1% | $1.125 |
| Gemini 3.5 Flash-Lite Google · Fast & low-cost | 1435 ±4 | 1453 ±7 | 26.0%high | 83.3%high | — | $0.85 |
| gpt-5.6-luna OpenAI | 1431 ±4xHigh | 1466 ±7xHigh | 82.1%max | 91.6%max | 41.0%max | $0.45 |
| Mistral Large 4 Mistral AI · Most capable | 1430 ±6 | 1478 ±11 | 58.6%high | 88.1%high | 20.0%high | $1.05 |
| grok-4.6 xAI | 1429 ±5High | 1464 ±7High | 66.0%xhigh | 94.0%high | 49.3%high | $3.00 |
| Beam 1.0 501B A23B Reflection AI | 1425 ±9 | 1474 ±18 | — | — | — | $0.60 |
| Claude Opus 4.1 Anthropic | 1419 ±3Thinking 16K, 20250805 | 1479 ±6Thinking 16K, 20250805 | — | — | — | $30.00 |
| Gemini 2.5 Flash Google | 1417 ±2 | 1423 ±4 | — | — | — | $0.85 |
| gpt-5.2 OpenAI | 1416 ±4High | 1442 ±6High | — | — | — | $4.8125 |
| gpt-5.4-mini OpenAI | 1412 ±4High | 1438 ±6High | — | — | — | $1.6875 |
| o3 OpenAI | 1410 ±42025-04-16 | 1407 ±62025-04-16 | — | — | — | $3.50 |
| gpt-5 OpenAI | 1406 ±4High | 1436 ±8High | — | — | — | $3.4375 |
| Claude Haiku 5.5 best value Anthropic · Fast & low-cost | 1401 ±8High | 1454 ±14High | 75.1%max | 89.6%max | 23.8%max | $0.20 |
| grok-4.7 xAI · Most capable | 1398 ±6xHigh | 1428 ±10xHigh | 53.0%xhigh | 92.7%xhigh | 56.0%xhigh | $3.00 |
| gpt-6-sol OpenAI | 1397 ±6Max | 1447 ±10Max | 89.8%max | 94.3%max | 60.7%max | $4.00 |
| grok-4.3 xAI | 1397 ±4 | 1414 ±6 | 42.8%high | 88.8%high | 33.2%high | $1.5625 |
| Claude Haiku 4.5 Anthropic | 1397 ±220251001 | 1454 ±420251001 | — | — | — | $2.00 |
| gpt-6-luna OpenAI · Fast & low-cost | 1389 ±4Max | 1437 ±6Max | 78.9%max | 90.5%max | 41.4%max | $0.20 |
| gpt-4.1 OpenAI | 1383 ±42025-04-14 | 1391 ±72025-04-14 | — | — | — | $3.50 |
| Claude Opus 4 Anthropic | 1377 ±4Thinking 16K, 20250514 | 1442 ±8Thinking 16K, 20250514 | — | — | — | $30.00 |
| gpt-5-mini OpenAI | 1373 ±5High | 1406 ±8High | — | — | — | $0.6875 |
| gpt-5.4-nano OpenAI | 1372 ±4High | 1404 ±6High | — | — | — | $0.4625 |
| o1 OpenAI | 1366 ±42024-12-17 | 1367 ±102024-12-17 | — | — | — | $26.25 |
| GPT OSS 120B OpenAI | 1365 ±4 | 1379 ±8 | — | 75.8%high | — | $0.2625 |
| o4-mini OpenAI | 1353 ±42025-04-16 | 1368 ±72025-04-16 | — | — | — | $1.925 |
| Claude Sonnet 4 Anthropic | 1351 ±4Thinking 32K, 20250514 | 1414 ±8Thinking 32K, 20250514 | — | — | — | $6.00 |
| gpt-4.1-mini OpenAI | 1340 ±42025-04-14 | 1367 ±82025-04-14 | — | — | — | $0.70 |
| o3-mini OpenAI | 1337 ±5High | 1378 ±12High | — | — | — | $1.925 |
| gpt-5-nano OpenAI | 1320 ±7High | 1351 ±15High | — | — | — | $0.1375 |
| gpt-4o OpenAI | 1300 ±32024-05-13 | 1297 ±62024-05-13 | — | — | — | $4.375 |
| GPT OSS 20B OpenAI | 1287 ±6 | 1306 ±13 | — | 60.8%medium | — | $0.13125 |
| gpt-4o-mini OpenAI | 1286 ±42024-07-18 | 1289 ±72024-07-18 | — | — | — | $0.2625 |
| gpt-4.1-nano OpenAI | 1285 ±82025-04-14 | 1305 ±192025-04-14 | — | — | — | $0.175 |
| gpt-3.5-turbo OpenAI | 1125 ±50125 | 1136 ±80125 | — | — | — | $0.75 |
| DeepSeek V4 Pro 0813 DeepSeek · Most capable | — | — | 64.6%max | 91.7%max | 52.9%max | $1.98 |
| DeepSeek V4.1 Flash DeepSeek | — | — | 67.4%max | 89.8%max | 42.9%max | $0.525 |
| DeepSeek V4.1 Flash 1017 DeepSeek · Fast & low-cost | — | — | 71.2%max | 93.2%max | 42.5%max | $0.525 |
| Gemini 3 Flash Preview Google | — | — | 51.2% | 89.4%high | 66.8%high | $1.125 |
| Gemini 3.1 Flash-Lite Google | — | — | 27.7%high | 81.8%high | — | $0.5625 |
| gpt-3.5-turbo-0125 OpenAI | — | — | 0.0% | 27.2% | — | $0.75 |
| gpt-4-0613 OpenAI | — | — | — | 30.7% | — | $37.50 |
| gpt-4-turbo-2024-04-09 OpenAI | — | — | 0.7% | 46.6% | — | $15.00 |
| gpt-4o-2024-05-13 OpenAI | — | — | — | 48.9% | — | $7.50 |
| gpt-5.5-pro OpenAI | — | — | 87.7%xhigh | — | — | $67.50 |
| gpt-6.1-terra OpenAI · Balanced | — | — | 90.9%max | 94.3%max | 57.3%max | $2.00 |
| grok-4.20-0309-reasoning xAI | — | — | 44.9% | 89.3% | 30.2% | $1.5625 |
| Muse Spark Open 816B A42B Meta | — | — | 69.8%xhigh | 88.6%xhigh | 56.6%xhigh | $2.00 |
| Qwen3.5 9B Qwen (Alibaba) | — | — | — | 79.0% | — | $0.19 |
| Qwen3.7 Max Qwen (Alibaba) · Most capable | — | — | 64.6% | 90.9% | 55.8% | $3.75 |
Each cell shows the model's best published result; the effort it was measured at is under the score (point at a cell for the date and other efforts). Models without any score are not listed: most small and open models are not tested by these sources.
What the benchmarks measure
- Arena text
- Rating from people's blind votes between two models' answers to their own prompts. It measures which answer people prefer, style included, not whether it is correct. Gaps inside the ± range are not meaningful.
- Arena coding
- The same blind votes, counted only for coding prompts.
- FrontierMath
- Unpublished, expert-written research-level math problems (tiers 1-3), run by Epoch AI. Share solved. Very hard: it separates the strongest reasoning models.
- GPQA Diamond
- 198 graduate-level multiple-choice questions in biology, physics and chemistry, run by Epoch AI. Share correct. Top models are close to the ceiling, so small gaps mean little.
- SimpleQA Verified
- Short fact questions answered without web search, run by Epoch AI. Share correct: how well the model recalls facts instead of guessing.
- Price / 1M
- The developer's standard API price per 1M tokens, weighted 3 input : 1 output (the usual way to compare one blended number). Reasoning models at higher effort write more hidden reasoning tokens, which are billed as output, so the same task costs more. Estimate your own workload on the models page.
Sources and licences
- LMArena: data from https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset, licensed CC BY 4.0. Fetched 2026-11-01; scores shown as published (best effort per model), matched to our models by name.
- Epoch AI: data from https://epoch.ai/benchmarks/use-this-data, licensed CC BY 4.0. Fetched 2026-11-01; scores shown as published (best effort per model), matched to our models by name. Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource].
Artificial Analysis scores are not shown: its terms do not allow reuse on a comparison site.