LLM benchmark comparison: price vs performance
Short answer: There is no single best LLM here: every benchmark measures a different skill. The highest LMArena Elo score is Claude Opus 5.5 at 1,526; one typical agent task on it costs $0.1168. The cheapest verified model per typical agent task is Gemini 2.5 Flash-Lite at $0.00277. Each score below keeps its own table, source and check date, and prices sit next to scores so you can see what one quality point costs.
Leaderboards by metric
Sorted by each metric separately; prices are the same verified list rates across every table.
AIME
36 of 55 verified models report this metric · unit: % · Higher is better · source: leaderboard
| # | Model | Score | Per agent task | Score checked | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Sol | 91.4% | $0.1207 | 2026-11-04 | source |
| 2 | Claude Opus 5 | 90.9% | $0.1509 | 2026-11-04 | source |
| 3 | Claude Sonnet 5 | 86.9% | $0.0604 | 2026-11-04 | source |
| 4 | DeepSeek V4 Pro | 86.8% | $0.0315 | 2026-11-04 | source |
| 5 | Gemini 3.6 Flash | 86.6% | $0.0226 | 2026-11-04 | source |
| 6 | Gemini 2.5 Pro | 85% | $0.0471 | 2026-11-04 | source |
| 7 | GPT-5.6 Luna | 82.2% | $0.00654 | 2026-11-04 | source |
| 8 | Gemini 3.5 Flash-Lite | 80.9% | $0.0116 | 2026-11-04 | source |
| 9 | GPT-6 Astra | 80.8% | $0.3018 | 2026-11-04 | source |
| 9 | GPT-5.6 Terra | 80.8% | $0.0654 | 2026-11-04 | source |
| 9 | GPT-5.5 | 80.8% | $0.1634 | 2026-11-04 | source |
| 9 | GPT-5.4 | 80.8% | $0.0817 | 2026-11-04 | source |
| 9 | Claude Fable 5.1 | 80.8% | $0.2873 | 2026-11-04 | source |
| 9 | Claude Sonnet 4.6 | 80.8% | $0.0905 | 2026-11-04 | source |
| 15 | GPT-6.1 Sol | 78.2% | $0.0584 | 2026-11-04 | source |
| 15 | GPT-6.1 Terra | 78.2% | $0.0292 | 2026-11-04 | source |
| 15 | GPT-6 Sol | 78.2% | $0.0604 | 2026-11-04 | source |
| 15 | GPT-5.4 mini | 78.2% | $0.0245 | 2026-11-04 | source |
| 15 | GPT-4.1 | 78.2% | $0.0611 | 2026-11-04 | source |
| 15 | o4-mini | 78.2% | $0.0336 | 2026-11-04 | source |
| 15 | Claude Sonnet 5.5 | 78.2% | $0.0584 | 2026-11-04 | source |
| 15 | Claude Haiku 4.5 | 78.2% | $0.0302 | 2026-11-04 | source |
| 15 | Gemini 4 Argon | 78.2% | $0.0584 | 2026-11-04 | source |
| 15 | Gemini 3.8 Flash | 78.2% | $0.0226 | 2026-11-04 | source |
| 25 | GPT-5.4 nano | 75.1% | $0.00666 | 2026-11-04 | source |
| 25 | GPT-5 mini | 75.1% | $0.00942 | 2026-11-04 | source |
| 25 | Gemini 3.1 Flash-Lite | 75.1% | $0.00817 | 2026-11-04 | source |
| 25 | Gemini 2.5 Flash | 75.1% | $0.0116 | 2026-11-04 | source |
| 25 | Mistral Large 4 | 75.1% | $0.0176 | 2026-11-04 | source |
| 30 | Claude Opus 5.5 | 74.8% | $0.1168 | 2026-11-04 | source |
| 31 | GPT-6 Luna | 71.5% | $0.00302 | 2026-11-04 | source |
| 31 | Claude Haiku 5.5 | 71.5% | $0.00302 | 2026-11-04 | source |
| 31 | Gemini 2.5 Flash-Lite | 71.5% | $0.00277 | 2026-11-04 | source |
| 31 | Mistral Small 4 | 71.5% | $0.00675 | 2026-11-04 | source |
| 31 | Codestral 25.08 | 71.5% | $0.0127 | 2026-11-04 | source |
| 36 | GPT-4o mini | 64.6% | $0.00531 | 2026-11-04 | source |
LiveCodeBench
36 of 55 verified models report this metric · unit: % · Higher is better · source: leaderboard
| # | Model | Score | Per agent task | Score checked | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Luna | 72.6% | $0.00654 | 2026-11-04 | source |
| 2 | GPT-6 Astra | 71.3% | $0.3018 | 2026-11-04 | source |
| 2 | GPT-5.6 Terra | 71.3% | $0.0654 | 2026-11-04 | source |
| 2 | GPT-5.5 | 71.3% | $0.1634 | 2026-11-04 | source |
| 2 | GPT-5.4 | 71.3% | $0.0817 | 2026-11-04 | source |
| 2 | Claude Fable 5.1 | 71.3% | $0.2873 | 2026-11-04 | source |
| 2 | Claude Sonnet 4.6 | 71.3% | $0.0905 | 2026-11-04 | source |
| 8 | Gemini 2.5 Pro | 70.9% | $0.0471 | 2026-11-04 | source |
| 9 | Claude Opus 5 | 69.7% | $0.1509 | 2026-11-04 | source |
| 10 | GPT-6.1 Sol | 69% | $0.0584 | 2026-11-04 | source |
| 10 | GPT-6.1 Terra | 69% | $0.0292 | 2026-11-04 | source |
| 10 | GPT-6 Sol | 69% | $0.0604 | 2026-11-04 | source |
| 10 | GPT-5.4 mini | 69% | $0.0245 | 2026-11-04 | source |
| 10 | GPT-4.1 | 69% | $0.0611 | 2026-11-04 | source |
| 10 | o4-mini | 69% | $0.0336 | 2026-11-04 | source |
| 10 | Claude Sonnet 5.5 | 69% | $0.0584 | 2026-11-04 | source |
| 10 | Claude Haiku 4.5 | 69% | $0.0302 | 2026-11-04 | source |
| 10 | Gemini 4 Argon | 69% | $0.0584 | 2026-11-04 | source |
| 10 | Gemini 3.8 Flash | 69% | $0.0226 | 2026-11-04 | source |
| 20 | GPT-5.6 Sol | 68.8% | $0.1207 | 2026-11-04 | source |
| 21 | Claude Sonnet 5 | 68.2% | $0.0604 | 2026-11-04 | source |
| 22 | Claude Opus 5.5 | 67.2% | $0.1168 | 2026-11-04 | source |
| 23 | DeepSeek V4 Pro | 66.4% | $0.0315 | 2026-11-04 | source |
| 24 | GPT-5.4 nano | 66.3% | $0.00666 | 2026-11-04 | source |
| 24 | GPT-5 mini | 66.3% | $0.00942 | 2026-11-04 | source |
| 24 | Gemini 3.1 Flash-Lite | 66.3% | $0.00817 | 2026-11-04 | source |
| 24 | Gemini 2.5 Flash | 66.3% | $0.0116 | 2026-11-04 | source |
| 24 | Mistral Large 4 | 66.3% | $0.0176 | 2026-11-04 | source |
| 29 | Gemini 3.6 Flash | 63.8% | $0.0226 | 2026-11-04 | source |
| 30 | GPT-6 Luna | 63.1% | $0.00302 | 2026-11-04 | source |
| 30 | Claude Haiku 5.5 | 63.1% | $0.00302 | 2026-11-04 | source |
| 30 | Gemini 2.5 Flash-Lite | 63.1% | $0.00277 | 2026-11-04 | source |
| 30 | Mistral Small 4 | 63.1% | $0.00675 | 2026-11-04 | source |
| 30 | Codestral 25.08 | 63.1% | $0.0127 | 2026-11-04 | source |
| 35 | Gemini 3.5 Flash-Lite | 60.7% | $0.0116 | 2026-11-04 | source |
| 36 | GPT-4o mini | 47.6% | $0.00531 | 2026-11-04 | source |
SWE-bench Verified
36 of 55 verified models report this metric · unit: % · Higher is better · source: leaderboard
| # | Model | Score | Per agent task | Score checked | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Luna | 69% | $0.00654 | 2026-11-04 | source |
| 2 | GPT-6 Astra | 67.8% | $0.3018 | 2026-11-04 | source |
| 2 | GPT-5.6 Terra | 67.8% | $0.0654 | 2026-11-04 | source |
| 2 | GPT-5.5 | 67.8% | $0.1634 | 2026-11-04 | source |
| 2 | GPT-5.4 | 67.8% | $0.0817 | 2026-11-04 | source |
| 2 | Claude Fable 5.1 | 67.8% | $0.2873 | 2026-11-04 | source |
| 2 | Claude Sonnet 4.6 | 67.8% | $0.0905 | 2026-11-04 | source |
| 8 | Gemini 2.5 Pro | 67.5% | $0.0471 | 2026-11-04 | source |
| 9 | Claude Opus 5 | 66.3% | $0.1509 | 2026-11-04 | source |
| 10 | GPT-6.1 Sol | 65.6% | $0.0584 | 2026-11-04 | source |
| 10 | GPT-6.1 Terra | 65.6% | $0.0292 | 2026-11-04 | source |
| 10 | GPT-6 Sol | 65.6% | $0.0604 | 2026-11-04 | source |
| 10 | GPT-5.4 mini | 65.6% | $0.0245 | 2026-11-04 | source |
| 10 | GPT-4.1 | 65.6% | $0.0611 | 2026-11-04 | source |
| 10 | o4-mini | 65.6% | $0.0336 | 2026-11-04 | source |
| 10 | Claude Sonnet 5.5 | 65.6% | $0.0584 | 2026-11-04 | source |
| 10 | Claude Haiku 4.5 | 65.6% | $0.0302 | 2026-11-04 | source |
| 10 | Gemini 4 Argon | 65.6% | $0.0584 | 2026-11-04 | source |
| 10 | Gemini 3.8 Flash | 65.6% | $0.0226 | 2026-11-04 | source |
| 20 | GPT-5.6 Sol | 65.4% | $0.1207 | 2026-11-04 | source |
| 21 | Claude Sonnet 5 | 64.9% | $0.0604 | 2026-11-04 | source |
| 22 | Claude Opus 5.5 | 63.9% | $0.1168 | 2026-11-04 | source |
| 23 | DeepSeek V4 Pro | 63.2% | $0.0315 | 2026-11-04 | source |
| 24 | GPT-5.4 nano | 63% | $0.00666 | 2026-11-04 | source |
| 24 | Gemini 3.1 Flash-Lite | 63% | $0.00817 | 2026-11-04 | source |
| 24 | Gemini 2.5 Flash | 63% | $0.0116 | 2026-11-04 | source |
| 24 | Mistral Large 4 | 63% | $0.0176 | 2026-11-04 | source |
| 28 | Gemini 3.6 Flash | 60.7% | $0.0226 | 2026-11-04 | source |
| 29 | GPT-6 Luna | 60.1% | $0.00302 | 2026-11-04 | source |
| 29 | Claude Haiku 5.5 | 60.1% | $0.00302 | 2026-11-04 | source |
| 29 | Gemini 2.5 Flash-Lite | 60.1% | $0.00277 | 2026-11-04 | source |
| 29 | Mistral Small 4 | 60.1% | $0.00675 | 2026-11-04 | source |
| 29 | Codestral 25.08 | 60.1% | $0.0127 | 2026-11-04 | source |
| 34 | Gemini 3.5 Flash-Lite | 57.7% | $0.0116 | 2026-11-04 | source |
| 35 | GPT-5 mini | 56.2% | $0.00942 | 2026-11-04 | source |
| 36 | GPT-4o mini | 45.2% | $0.00531 | 2026-11-04 | source |
LMArena Elo
33 of 55 verified models report this metric · unit: Elo · Higher is better · source: leaderboard
| # | Model | Score | Per agent task | Score checked | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 1,526 | $0.1168 | 2026-11-04 | source |
| 2 | GPT-6 Astra | 1,525 | $0.3018 | 2026-11-04 | source |
| 3 | Gemini 4 Argon | 1,523 | $0.0584 | 2026-11-04 | source |
| 4 | Claude Sonnet 5.5 | 1,521 | $0.0584 | 2026-11-04 | source |
| 5 | Claude Fable 5.1 | 1,520 | $0.2873 | 2026-11-04 | source |
| 6 | GPT-6.1 Sol | 1,516 | $0.0584 | 2026-11-04 | source |
| 7 | Claude Opus 5 | 1,512 | $0.1509 | 2026-11-04 | source |
| 8 | GPT-6 Sol | 1,509 | $0.0604 | 2026-11-04 | source |
| 9 | GPT-5.6 Sol | 1,508 | $0.1207 | 2026-11-04 | source |
| 10 | Grok 4.7 | 1,507 | $0.0561 | 2026-11-04 | source |
| 11 | GPT-5.6 Terra | 1,505 | $0.0654 | 2026-11-04 | source |
| 11 | GPT-5.5 | 1,505 | $0.1634 | 2026-11-04 | source |
| 11 | Gemini 3.8 Flash | 1,505 | $0.0226 | 2026-11-04 | source |
| 14 | Gemini 3.1 Pro Preview | 1,504 | $0.0654 | 2026-11-04 | source |
| 15 | DeepSeek Flash (V4.1) | 1,503 | $0.00784 | 2026-11-04 | source |
| 16 | DeepSeek V4 Pro | 1,502 | $0.0315 | 2026-11-04 | source |
| 17 | Claude Haiku 5.5 | 1,501 | $0.00302 | 2026-11-04 | source |
| 17 | Gemini 3.6 Flash | 1,501 | $0.0226 | 2026-11-04 | source |
| 19 | Claude Sonnet 5 | 1,485 | $0.0604 | 2026-11-04 | source |
| 20 | GPT-5.4 | 1,465 | $0.0817 | 2026-11-04 | source |
| 21 | GPT-5.4 mini | 1,461 | $0.0245 | 2026-11-04 | source |
| 22 | Claude Sonnet 4.6 | 1,460 | $0.0905 | 2026-11-04 | source |
| 23 | Gemini 2.5 Pro | 1,459 | $0.0471 | 2026-11-04 | source |
| 24 | GPT-5.6 Luna | 1,451 | $0.00654 | 2026-11-04 | source |
| 25 | Mistral Large 4 | 1,448 | $0.0176 | 2026-11-04 | source |
| 26 | Gemini 3.1 Flash-Lite | 1,421 | $0.00817 | 2026-11-04 | source |
| 27 | Gemini 2.5 Flash | 1,412 | $0.0116 | 2026-11-04 | source |
| 28 | GPT-4.1 | 1,381 | $0.0611 | 2026-11-04 | source |
| 29 | Claude Haiku 4.5 | 1,378 | $0.0302 | 2026-11-04 | source |
| 30 | GPT-5 mini | 1,375 | $0.00942 | 2026-11-04 | source |
| 31 | o4-mini | 1,362 | $0.0336 | 2026-11-04 | source |
| 32 | Gemini 2.5 Flash-Lite | 1,337 | $0.00277 | 2026-11-04 | source |
| 33 | GPT-4o mini | 1,289 | $0.00531 | 2026-11-04 | source |
Humanity's Last Exam
24 of 55 verified models report this metric · unit: % · Higher is better · source: leaderboard
| # | Model | Score | Per agent task | Score checked | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 61.4% | $0.1168 | 2026-11-04 | source |
| 2 | Claude Fable 5.1 | 59.1% | $0.2873 | 2026-11-04 | source |
| 3 | Gemini 4 Argon | 57.1% | $0.0584 | 2026-11-04 | source |
| 4 | Claude Sonnet 5.5 | 55% | $0.0584 | 2026-11-04 | source |
| 5 | Claude Opus 5 | 54.9% | $0.1509 | 2026-11-04 | source |
| 6 | GPT-6 Astra | 54.7% | $0.3018 | 2026-11-04 | source |
| 7 | GPT-6.1 Sol | 52.9% | $0.0584 | 2026-11-04 | source |
| 8 | GPT-5.6 Sol | 49.5% | $0.1207 | 2026-11-04 | source |
| 9 | GPT-6 Sol | 47.9% | $0.0604 | 2026-11-04 | source |
| 10 | Gemini 3.8 Flash | 47.8% | $0.0226 | 2026-11-04 | source |
| 11 | Gemini 3.1 Pro Preview | 47% | $0.0654 | 2026-11-04 | source |
| 12 | GPT-6.1 Terra | 46.2% | $0.0292 | 2026-11-04 | source |
| 13 | GPT-5.5 | 45.8% | $0.1634 | 2026-11-04 | source |
| 14 | Claude Haiku 5.5 | 44.4% | $0.00302 | 2026-11-04 | source |
| 15 | Grok 4.7 | 43.1% | $0.0561 | 2026-11-04 | source |
| 16 | GPT-5.6 Terra | 42.9% | $0.0654 | 2026-11-04 | source |
| 17 | Gemini 3.6 Flash | 40.8% | $0.0226 | 2026-11-04 | source |
| 18 | GPT-6 Luna | 38.5% | $0.00302 | 2026-11-04 | source |
| 19 | Gemini 2.5 Pro | 22.5% | $0.0471 | 2026-11-04 | source |
| 20 | Gemini 3.5 Flash-Lite | 18.8% | $0.0116 | 2026-11-04 | source |
| 21 | Ministral 3 3B | 5.4% | $0.00375 | 2026-11-04 | source |
| 22 | Ministral 3 14B | 4.6% | $0.0075 | 2026-11-04 | source |
| 23 | Ministral 3 8B | 4.3% | $0.00562 | 2026-11-04 | source |
| 24 | Mistral Small 4 | 3.8% | $0.00675 | 2026-11-04 | source |
Output speed
24 of 55 verified models report this metric · unit: tok/s · Higher is better · source: leaderboard
| # | Model | Score | Per agent task | Score checked | Source |
|---|---|---|---|---|---|
| 1 | Gemini 3.8 Flash | 345 | $0.0226 | 2026-11-04 | source |
| 2 | DeepSeek Flash (V4.1) | 217 | $0.00784 | 2026-11-04 | source |
| 3 | GPT-5.6 Luna | 186 | $0.00654 | 2026-11-04 | source |
| 4 | GPT-6 Sol | 153 | $0.0604 | 2026-11-04 | source |
| 5 | Gemini 3.6 Flash | 148 | $0.0226 | 2026-11-04 | source |
| 6 | Claude Sonnet 5.5 | 139 | $0.0584 | 2026-11-04 | source |
| 7 | Gemini 3.1 Pro Preview | 131 | $0.0654 | 2026-11-04 | source |
| 8 | GPT-4.1 | 120 | $0.0611 | 2026-11-04 | source |
| 9 | GPT-5.6 Terra | 118 | $0.0654 | 2026-11-04 | source |
| 10 | GPT-6.1 Terra | 109 | $0.0292 | 2026-11-04 | source |
| 11 | Claude Sonnet 5 | 98 | $0.0604 | 2026-11-04 | source |
| 12 | GPT-5.6 Sol | 96 | $0.1207 | 2026-11-04 | source |
| 13 | Claude Opus 5.5 | 94 | $0.1168 | 2026-11-04 | source |
| 14 | GPT-5.4 | 88 | $0.0817 | 2026-11-04 | source |
| 15 | Gemini 2.5 Pro | 87 | $0.0471 | 2026-11-04 | source |
| 16 | Grok 4.3 | 83 | $0.0298 | 2026-11-04 | source |
| 17 | Gemini 4 Argon | 82 | $0.0584 | 2026-11-04 | source |
| 18 | GPT-6.1 Sol | 74 | $0.0584 | 2026-11-04 | source |
| 18 | Claude Opus 5 | 74 | $0.1509 | 2026-11-04 | source |
| 20 | Claude Sonnet 4.6 | 73 | $0.0905 | 2026-11-04 | source |
| 21 | GPT-5.5 | 70 | $0.1634 | 2026-11-04 | source |
| 22 | Claude Fable 5.1 | 68 | $0.2873 | 2026-11-04 | source |
| 23 | DeepSeek V4 Pro | 62 | $0.0315 | 2026-11-04 | source |
| 24 | GPT-6 Astra | 53 | $0.3018 | 2026-11-04 | source |
MMMU-Pro
20 of 55 verified models report this metric · unit: % · Higher is better · source: leaderboard
| # | Model | Score | Per agent task | Score checked | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 87.7% | $0.1168 | 2026-11-04 | source |
| 2 | GPT-6 Astra | 86.9% | $0.3018 | 2026-11-04 | source |
| 3 | GPT-6.1 Sol | 86% | $0.0584 | 2026-11-04 | source |
| 4 | Gemini 3.8 Flash | 85.6% | $0.0226 | 2026-11-04 | source |
| 5 | GPT-6.1 Terra | 84.7% | $0.0292 | 2026-11-04 | source |
| 5 | Claude Opus 5 | 84.7% | $0.1509 | 2026-11-04 | source |
| 7 | GPT-5.6 Sol | 83.4% | $0.1207 | 2026-11-04 | source |
| 8 | Gemini 3.6 Flash | 83.2% | $0.0226 | 2026-11-04 | source |
| 9 | GPT-6 Sol | 82.9% | $0.0604 | 2026-11-04 | source |
| 10 | Gemini 3.1 Pro Preview | 82.4% | $0.0654 | 2026-11-04 | source |
| 11 | GPT-5.6 Terra | 80.7% | $0.0654 | 2026-11-04 | source |
| 12 | GPT-5.5 | 79.9% | $0.1634 | 2026-11-04 | source |
| 13 | GPT-6 Luna | 79.7% | $0.00302 | 2026-11-04 | source |
| 14 | Gemini 3.5 Flash-Lite | 79% | $0.0116 | 2026-11-04 | source |
| 15 | Grok 4.7 | 78.2% | $0.0561 | 2026-11-04 | source |
| 16 | Gemini 2.5 Pro | 74.9% | $0.0471 | 2026-11-04 | source |
| 17 | Ministral 3 14B | 49.8% | $0.0075 | 2026-11-04 | source |
| 18 | Mistral Small 4 | 46.5% | $0.00675 | 2026-11-04 | source |
| 19 | Ministral 3 8B | 46% | $0.00562 | 2026-11-04 | source |
| 20 | Ministral 3 3B | 38.1% | $0.00375 | 2026-11-04 | source |
GPQA Diamond
15 of 55 verified models report this metric · unit: % · Higher is better · source: leaderboard
| # | Model | Score | Per agent task | Score checked | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 96.1% | $0.3018 | 2026-11-04 | source |
| 2 | Gemini 3.8 Flash | 95.3% | $0.0226 | 2026-11-04 | source |
| 3 | GPT-5.6 Sol | 94.1% | $0.1207 | 2026-11-04 | source |
| 3 | Gemini 3.1 Pro Preview | 94.1% | $0.0654 | 2026-11-04 | source |
| 5 | Claude Fable 5.1 | 93.7% | $0.2873 | 2026-11-04 | source |
| 6 | GPT-5.5 | 93.5% | $0.1634 | 2026-11-04 | source |
| 7 | Claude Opus 5 | 93.2% | $0.1509 | 2026-11-04 | source |
| 8 | Gemini 3.6 Flash | 92.8% | $0.0226 | 2026-11-04 | source |
| 9 | GPT-5.6 Terra | 92.5% | $0.0654 | 2026-11-04 | source |
| 10 | Gemini 2.5 Pro | 84.4% | $0.0471 | 2026-11-04 | source |
| 11 | Gemini 3.5 Flash-Lite | 83.8% | $0.0116 | 2026-11-04 | source |
| 12 | Ministral 3 14B | 57.2% | $0.0075 | 2026-11-04 | source |
| 13 | Mistral Small 4 | 57.1% | $0.00675 | 2026-11-04 | source |
| 14 | Ministral 3 8B | 47.1% | $0.00562 | 2026-11-04 | source |
| 15 | Ministral 3 3B | 35.8% | $0.00375 | 2026-11-04 | source |
Price vs performance
Reading the chart
Cheaper models sit to the left, higher-scoring models sit higher. Points in the lower-left are cheap and weak, points in the upper-right are expensive and strong, and the interesting ones sit above the trend: high score at a low task cost.
Cost per quality point
| Metric | Models | Best value | Lowest $/point | Median $/point | Highest $/point |
|---|---|---|---|---|---|
| AIME | 36 | Gemini 2.5 Flash-Lite | $0.0000387 | $0.000380 | $0.003735 |
| LiveCodeBench | 36 | Gemini 2.5 Flash-Lite | $0.0000439 | $0.000456 | $0.004232 |
| SWE-bench Verified | 36 | Gemini 2.5 Flash-Lite | $0.0000460 | $0.000479 | $0.004451 |
| Humanity's Last Exam | 24 | Claude Haiku 5.5 | $0.0000680 | $0.001305 | $0.005516 |
| MMMU-Pro | 20 | GPT-6 Luna | $0.0000379 | $0.000654 | $0.003472 |
| GPQA Diamond | 15 | Ministral 3 3B | $0.000105 | $0.000558 | $0.003140 |
Cost per point = the price of one typical agent task on the model divided by its score on that metric. It only works on scales that start at zero and reach 100, so it is not shown for Elo ratings or latency. Read it as "what one point of this score costs", not as a quality score of its own: a model can be the best value on one metric and mediocre on another.
What the ratio hides
A low cost per point can come from a cheap model with a weak score or from an expensive model near the top. Both can be the right buy depending on the bottleneck, so use the ratio to shortlist two or three models and then run your own prompts on them.
Method and sources
One table per metric
We never add two benchmarks into one combined score. A percentage, an Elo rating and a time are different scales, and averaging them would hide which one moved. Each table is sorted by its own metric only, and the column header says which direction is better.
Missing is missing
If a model has no published value for a metric it does not appear in that table. It is not shown as zero, it is not estimated from a sibling model, and it does not affect the ranks of the models that do report the metric.
Ties and ranks
Models with the same score share a rank; the next rank skips the tied places. Ranks are within the set of models that report that metric, not across the whole price list.
What "one agent task" means for the price column
Every price column prices the same task: 35K input tokens (55% cached) and 2,500 output tokens per task. That basis is median "typical task" token counts across the 14 tools in src/tools.json (35K input / 2,500 output, 55% cached). Rates are the standard list prices in the price table; batch and long-context tiers are not applied.
Where the scores come from
| Metric | Unit | Better | Models reporting | Source | Data checked |
|---|---|---|---|---|---|
| AIME | % | higher | 36 | www.trustthebench.com | 2026-11-04 |
| LiveCodeBench | % | higher | 36 | www.trustthebench.com | 2026-11-04 |
| SWE-bench Verified | % | higher | 36 | www.trustthebench.com | 2026-11-04 |
| LMArena Elo | Elo | higher | 33 | openlm.ai | 2026-11-04 |
| Humanity's Last Exam | % | higher | 24 | artificialanalysis.ai | 2026-11-04 |
| Output speed | tok/s | higher | 24 | www.swfte.com | 2026-11-04 |
| MMMU-Pro | % | higher | 20 | artificialanalysis.ai | 2026-11-04 |
| GPQA Diamond | % | higher | 15 | artificialanalysis.ai | 2026-11-04 |
Benchmark data compiled 2026-11-04; each model stores the actual read date in 'checked'. SWE-bench Verified, LiveCodeBench and AIME values are individual per-model rows from TrustTheBench (an aggregator that re-publishes standardized-protocol results, snapshot 2026-09-12, each row linking to the original benchmark source). TrustTheBench is tier-clustered: its 2026-09-12 snapshot reports the same rounded score for several models (e.g. 65.6 on SWE-bench Verified), so equal scores mean 'same reported tier', not identical capability, and small score differences are not meaningful. GPQA Diamond, MMMU-Pro and Humanity's Last Exam come from Artificial Analysis, which runs those evaluations itself; where a model publishes several reasoning-effort variants we use the highest-effort published variant and say so in the model 'notes'. LMArena Elo comes from the public Chatbot Arena+ leaderboard at openlm.ai (an aggregator snapshot, not the live lmarena.ai API); where that leaderboard lists a dated snapshot (e.g. GPT-4.1-2025-04-14) or a reasoning variant (e.g. GPT-5.5-high) for the same API model, we use it and record the exact entry in the model 'notes'. Output speed is the 'Speed' column of a public LMArena mirror (swfte.com/lmarena, aggregator), with a few values not present there taken from the BenchLeader coding leaderboard (per-model 'sources' override). Time-to-first-token and Aider Polyglot / BFCL are intentionally not published here: the accessible sources did not provide reliable per-model values (Artificial Analysis' latency figure for highest-effort reasoning variants is time-to-first-answer-token, not TTFT), and a missing metric is omitted rather than estimated, extrapolated or copied from a model family. Benchmarks are never comparable with each other - rank within one metric only.
Scores shared across models
Some metrics repeat the same rounded value across models: AIME: 4 repeated values covering 26 of 36 rows; LMArena Elo: 2 repeated values covering 5 of 33 rows; GPQA Diamond: 1 repeated value covering 2 of 15 rows; LiveCodeBench: 4 repeated values covering 26 of 36 rows; MMMU-Pro: 1 repeated value covering 2 of 20 rows; SWE-bench Verified: 4 repeated values covering 25 of 36 rows. An equal value means the source published the same reported figure, not that the models are equally capable on that metric, so we do not use these ties for fine-grained ranking between the models that share one.
For SWE-bench Verified, LiveCodeBench and AIME the values come from TrustTheBench, an aggregator that republishes results under a standardized protocol (mini-SWE-agent 2.0.0, snapshot 2026-09-12). Its published snapshot is tier-clustered: leading results are reported in the same band (its own page lists several models at 75.8%), so identical scores above indicate the same reporting band in that snapshot. We keep each model's own row from the source, but treat a shared value as a tie at the source rather than as evidence of equal capability.
Repeated LMArena Elo values come from the source's rounded ratings: the same Elo in the table is a tie in the published snapshot, not a separate measurement.
What this page does not claim
Leaderboards are not your product. A model that wins a public benchmark can still fail on your prompt, your data and your tool schema; treat the tables as a shortlist and run a small eval on real examples before you switch. Prices change, and the check date on each row is the only promise we make about a number.
Related tools, skills & plugins
Benchmarks tell you which model to shortlist; the token bill is decided by the agent, skill or plugin you run on it. These entries have a cost page or an official site, and the hub lists every entry by category.
- Claude Code — Anthropic's terminal coding agent for multi-step software tasks. ($0.0810 per task, Documented)
- OpenAI Codex CLI — OpenAI's open-source terminal coding agent, billed via ChatGPT credits. ($0.0722 per task, Documented)
- Cursor — AI-first code editor and agent with per-token third-party model billing. ($0.0760 per task, Estimated)
- GitHub Copilot coding agent — GitHub's asynchronous coding agent, billed with AI Credits from token use. ($0.1180 per task, Estimated)
- All skills, plugins and MCP token costs
Related
- Best LLM for coding: benchmarks and cost per task
- Compare two to four models side by side
- Every model price per 1M tokens
- AI design leaderboard: Gestalt vs Claude Fable 5.2
- Find a model for your use case
- What one task costs in tokens
Questions
Which LLM is best overall?
There is no overall winner because the benchmarks measure different skills. The highest LMArena Elo score is Claude Opus 5.5 at 1,526, while the cheapest model per typical agent task is Gemini 2.5 Flash-Lite at $0.00277. Pick the metric that matches your workload and then compare cost at your own token counts.
Why not publish one combined benchmark score?
Because a combined score needs weights, and any weighting hides the trade-off you are buying. Keeping one table per metric lets you apply your own weights: a coding agent and a document summarizer do not need the same ranking.
Where do the benchmark numbers and dates come from?
Each metric names its source in the source table at the bottom of this page and every row shows the date that score was checked. Primary leaderboards are used first; when a figure comes from an aggregator that is stated in the note.
Why is a model missing from one of the tables?
That model does not report the metric, so it is left out rather than shown as zero or estimated. The table header shows how many of the verified models report each metric.
How do you compute cost per quality point?
Divide the list-price cost of one typical agent task on the model by its score on a 0-100 metric. It is only shown for metrics with a zero-to-100 scale; Elo ratings and latency are excluded because a per-point ratio would not mean anything.
Are these scores independent measurements of every model?
Not always. Some sources repeat the same rounded score across models: the SWE-bench, LiveCodeBench and AIME values here come from a standardized-protocol snapshot that is tier-clustered, so several models can sit in the same reported band. The shared-scores note under Method lists every metric and row count affected; treat a repeated value as a tie at the source, not as evidence of equal capability or a ranking gap.
How current is this page?
The benchmark data was last read on 2026-11-04 and each row keeps its own check date; prices were last checked on 2026-11-18. When a source changes a score we update the row and move its date, and a score without a readable source is removed rather than kept as a stale number.
Next step → See what one task costs in tokens
Price data last checked 2026-11-18 against official sources · how we verify
Price-change alerts
One short email when a provider changes a price or a promo ends. No spam, unsubscribe any time.