NanoGPT Speedrun Frontier
Agents change a GPT optimizer on fixed GPUs. Each candidate must reach the target validation loss across eight fixed seeds using fewer training steps.
Prime Intellect ran 153 autonomous trials across 18 frontier models and published model trajectories, costs and verified records.
- Made by
- Prime Intellect
- Released
- August 2026
- Kind
- Machine learning engineering
- Level
- Research frontier
- Scored
- Checked automatically
- Answers
- Agent task
- Measures
- Efficiency
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal
- Refresh
- One fixed set
- Run by
- A host
- In views
- Agents
- Strongest model tested
- Claude Opus 5 #11 on the capability index
- Last model added
- August 2026
- Scores refreshed
- November 21st, 2026
Claude Fable 5 leads with 2726.0 among models released by August 2026 Claude Fable 5 leads with 2726.0 among models released by August 2026
Best score per entry from the benchmark's own leaderboard, read on November 21st, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in August 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 12 more
- 1 Claude Fable 5 high, c, subagent 2726.0
- 2 Claude Opus 5 max, a, serial 2920.0
- 3 Kimi K3 · prime-agent max, a, serial 2930.0
- 4 Kimi K3 max, a, subagent 2974.0
- 5 Claude Opus 4.8 max, a, subagent 3018.0
- 6 GPT-5.6 Sol xhigh, a, subagent 3042.0
- 7 GPT-5.6 Sol Pro xhigh, a, serial 3058.0
- 8 Claude Sonnet 5 max, a, subagent 3105.0
- 9 GPT-5.6 Luna xhigh, a, subagent 3110.0
- 10 Grok 4.5 xhigh, c, subagent 3120.0
- 11
Qwen 3.8 Max max, b, serial 3120.0
- 12 GLM-5.2 high, a, subagent 3150.0
- 13
DeepSeek-V4-Pro max, c, serial 3205.0
- 14 GPT-5.6 Terra xhigh, a, serial 3214.0
- 15 Grok 4.6 xhigh, a, subagent 3220.0
- 16
Muse Spark 1.2 xhigh, a, subagent 3230.0
- 17
Muse Spark 1.1 max, a, subagent 3232.0
- 18 GPT-5.5 xhigh, a, serial 3234.0
- 19 Kimi K2.7 Code max, a, subagent 3240.0
Source: Official NanoGPT Speedrun Frontier leaderboard
- 1 Claude Fable 5 high, c, subagent 2726.0
- 2 Claude Opus 5 max, a, serial 2920.0
- 3 Kimi K3 max, a, subagent 2974.0
- 4 Claude Opus 4.8 max, a, subagent 3018.0
- 5 GPT-5.6 Sol xhigh, a, subagent 3042.0
- 6 GPT-5.6 Sol Pro xhigh, a, serial 3058.0
- 7 Claude Sonnet 5 max, a, subagent 3105.0
- 8 GPT-5.6 Luna xhigh, a, subagent 3110.0
- 9 Grok 4.5 xhigh, c, subagent 3120.0
- 10
Qwen 3.8 Max max, b, serial 3120.0
- 11 GLM-5.2 high, a, subagent 3150.0
- 12
DeepSeek-V4-Pro max, c, serial 3205.0
- 13 GPT-5.6 Terra xhigh, a, serial 3214.0
- 14 Grok 4.6 xhigh, a, subagent 3220.0
- 15
Muse Spark 1.2 xhigh, a, subagent 3230.0
- 16
Muse Spark 1.1 max, a, subagent 3232.0
- 17 GPT-5.5 xhigh, a, serial 3234.0
- 18 Kimi K2.7 Code max, a, subagent 3240.0
Source: Official NanoGPT Speedrun Frontier leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Fable 5 | 2726.0 | high, c, subagent, claude-code | Jun 2026 | Official leaderboard |
| 2 | Claude Opus 5 | 2920.0 | max, a, serial, claude-code | Jul 2026 | Official leaderboard |
| 3 | Kimi K3 · prime-agent | 2930.0 | max, a, serial, prime-agent | Official leaderboard | |
| 4 | Kimi K3 | 2974.0 | max, a, subagent, kimi-code | Jul 2026 | Official leaderboard |
| 5 | Claude Opus 4.8 | 3018.0 | max, a, subagent, claude-code | May 2026 | Official leaderboard |
| 6 | GPT-5.6 Sol | 3042.0 | xhigh, a, subagent, codex | Jul 2026 | Official leaderboard |
| 7 | GPT-5.6 Sol Pro | 3058.0 | xhigh, a, serial, codex | Jul 2026 | Official leaderboard |
| 8 | Claude Sonnet 5 | 3105.0 | max, a, subagent, claude-code | Jun 2026 | Official leaderboard |
| 9 | GPT-5.6 Luna | 3110.0 | xhigh, a, subagent, codex | Jul 2026 | Official leaderboard |
| 10 | Grok 4.5 | 3120.0 | xhigh, c, subagent, grok-cli | Jul 2026 | Official leaderboard |
| 11 | | 3120.0 | max, b, serial, qwen-code | Aug 2026 | Official leaderboard |
| 12 | GLM-5.2 | 3150.0 | high, a, subagent, pi | Jun 2026 | Official leaderboard |
| 13 | | 3205.0 | max, c, serial, claude-code | Apr 2026 | Official leaderboard |
| 14 | GPT-5.6 Terra | 3214.0 | xhigh, a, serial, codex | Jul 2026 | Official leaderboard |
| 15 | Grok 4.6 | 3220.0 | xhigh, a, subagent, grok-cli-sub | Aug 2026 | Official leaderboard |
| 16 | | 3230.0 | xhigh, a, subagent, muse-code-sub | Aug 2026 | Official leaderboard |
| 17 | | 3232.0 | max, a, subagent, pi | Jul 2026 | Official leaderboard |
| 18 | GPT-5.5 | 3234.0 | xhigh, a, serial, codex | Apr 2026 | Official leaderboard |
| 19 | Kimi K2.7 Code | 3240.0 | max, a, subagent, kimi-code | Jun 2026 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Fable 5 | 2726.0 | high, c, subagent, claude-code | Jun 2026 | Official leaderboard |
| 2 | Claude Opus 5 | 2920.0 | max, a, serial, claude-code | Jul 2026 | Official leaderboard |
| 3 | Kimi K3 | 2974.0 | max, a, subagent, kimi-code | Jul 2026 | Official leaderboard |
| 4 | Claude Opus 4.8 | 3018.0 | max, a, subagent, claude-code | May 2026 | Official leaderboard |
| 5 | GPT-5.6 Sol | 3042.0 | xhigh, a, subagent, codex | Jul 2026 | Official leaderboard |
| 6 | GPT-5.6 Sol Pro | 3058.0 | xhigh, a, serial, codex | Jul 2026 | Official leaderboard |
| 7 | Claude Sonnet 5 | 3105.0 | max, a, subagent, claude-code | Jun 2026 | Official leaderboard |
| 8 | GPT-5.6 Luna | 3110.0 | xhigh, a, subagent, codex | Jul 2026 | Official leaderboard |
| 9 | Grok 4.5 | 3120.0 | xhigh, c, subagent, grok-cli | Jul 2026 | Official leaderboard |
| 10 | | 3120.0 | max, b, serial, qwen-code | Aug 2026 | Official leaderboard |
| 11 | GLM-5.2 | 3150.0 | high, a, subagent, pi | Jun 2026 | Official leaderboard |
| 12 | | 3205.0 | max, c, serial, claude-code | Apr 2026 | Official leaderboard |
| 13 | GPT-5.6 Terra | 3214.0 | xhigh, a, serial, codex | Jul 2026 | Official leaderboard |
| 14 | Grok 4.6 | 3220.0 | xhigh, a, subagent, grok-cli-sub | Aug 2026 | Official leaderboard |
| 15 | | 3230.0 | xhigh, a, subagent, muse-code-sub | Aug 2026 | Official leaderboard |
| 16 | | 3232.0 | max, a, subagent, pi | Jul 2026 | Official leaderboard |
| 17 | GPT-5.5 | 3234.0 | xhigh, a, serial, codex | Apr 2026 | Official leaderboard |
| 18 | Kimi K2.7 Code | 3240.0 | max, a, subagent, kimi-code | Jun 2026 | Official leaderboard |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official NanoGPT Speedrun Frontier leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| max, b, subagent | claude-code-sub | 3288.0 | 1,256,275 | ||
| xhigh, c, serial | claude-code | 3280.0 | $195 | 949,721 | |
| GPT-5.5 | xhigh, c, serial | codex | 3270.0 | $262 | 164,546 |
| max, a, subagent | prime-agent-sub | 3270.0 | 456,743 | ||
| Grok 4.5 | xhigh, a, serial | grok-cli-goal | 3270.0 | $112 | 110,871 |
| Grok 4.5 | xhigh, b, serial | grok-cli-goal | 3270.0 | $150 | 162,676 |
| xhigh, b, subagent | muse-code-sub | 3270.0 | 1,373,908 | ||
| max, a, serial | claude-code | 3265.0 | 331,253 | ||
| max, d, serial | claude-code | 3263.0 | 219,121 | ||
| max, a, subagent | claude-code-sub | 3263.0 | 703,172 | ||
| xhigh, c, serial | muse-code | 3263.0 | 2,362,863 | ||
| GLM-5.2 | xhigh, d, serial | claude-code | 3262.0 | $30 | 430,697 |
| Kimi K2.7 Code | max, b, subagent | kimi-code | 3260.0 | $34 | 727,053 |
| Kimi K2.7 Code | high, a, subagent | pi | 3260.0 | $20 | 294,328 |
| Claude Sonnet 5 | xhigh, b, subagent | claude-code | 3260.0 | $21 | 207,800 |
| xhigh, a, serial | claude-code-goal | 3260.0 | $97 | 675,416 | |
| Grok 4.6 | xhigh, b, subagent | grok-cli-sub | 3255.0 | 257,781 | |
| Kimi K2.7 Code | max, d, serial | kimi-code | 3251.0 | $6.01 | 1,501,977 |
| max, a, serial | prime-agent | 3250.0 | 1,820,857 | ||
| Claude Opus 4.8 | xhigh, b, subagent | claude-code | 3250.0 | $75 | 343,451 |
| xhigh, a, subagent | claude-code | 3250.0 | $324 | 1,343,151 | |
| Claude Opus 5 | max, b, serial | claude-code | 3250.0 | $51 | 210,171 |
| max, a, subagent | claude-code-sub | 3250.0 | 609,916 | ||
| GPT-5.6 Luna | xhigh, c, subagent | codex | 3247.0 | $21 | 859,634 |
| GPT-5.6 Terra | xhigh, c, serial | codex | 3245.0 | $84 | 219,454 |
| GPT-5.5 | xhigh, b, serial | codex | 3243.0 | $251 | 157,766 |
| GLM-5.2 | xhigh, a, serial | claude-code | 3243.0 | $27 | 385,518 |
| Grok 4.6 | xhigh, b, serial | grok-cli | 3243.0 | 287,723 | |
| GPT-5.6 Terra | xhigh, b, serial | codex | 3242.0 | $47 | 186,103 |
| Kimi K2.7 Code | max, a, subagent | kimi-code | 3240.0 | $51 | 763,161 |
| GLM-5.2 | xhigh, b, serial | claude-code | 3240.0 | $14 | 249,251 |
| max, b, subagent | pi | 3240.0 | $20 | 1,528,524 | |
| max, b, subagent | claude-code-sub | 3240.0 | 488,662 | ||
| xhigh, b, serial | muse-code | 3240.0 | 2,572,145 | ||
| Grok 4.6 | xhigh, c, serial | grok-cli | 3240.0 | 279,394 | |
| Claude Opus 4.8 | max, b, subagent | claude-code | 3236.0 | $366 | 1,466,322 |
| GPT-5.5 | xhigh, a, serial | codex | 3234.0 | $158 | 77,425 |
| max, a, subagent | pi | 3232.0 | $31 | 1,603,714 | |
| Claude Sonnet 5 | xhigh, a, subagent | claude-code | 3230.0 | $21 | 205,476 |
| Kimi K3 | max, a, serial | claude-code-goal | 3230.0 | $101 | 1,081,557 |
| xhigh, a, subagent | muse-code-sub | 3230.0 | 910,087 | ||
| Grok 4.6 | xhigh, a, serial | grok-cli | 3227.0 | 305,686 | |
| Claude Opus 4.8 | xhigh, a, subagent | claude-code | 3226.0 | $75 | 470,614 |
| max, b, serial | prime-agent | 3225.0 | 2,613,844 | ||
| Grok 4.6 | xhigh, a, subagent | grok-cli-sub | 3220.0 | 345,731 | |
| Kimi K3 | max, b, subagent | kimi-code | 3219.0 | $28 | 381,742 |
| GLM-5.2 | xhigh, c, subagent | claude-code | 3217.0 | $42 | 530,685 |
| GPT-5.6 Terra | xhigh, a, serial | codex | 3214.0 | $92 | 297,651 |
| max, c, serial | claude-code | 3205.0 | 319,065 | ||
| GLM-5.2 | xhigh, a, subagent | claude-code | 3205.0 | $30 | 536,371 |
| GLM-5.2 | xhigh, a, serial | claude-code-goal | 3201.0 | $68 | 548,989 |
| GLM-5.2 | xhigh, c, serial | claude-code | 3190.0 | $21 | 316,234 |
| GPT-5.6 Luna | xhigh, b, subagent | codex | 3185.0 | $15 | 730,250 |
| GPT-5.6 Sol | xhigh, b, subagent | codex | 3184.0 | $218 | 410,516 |
| max, c, serial | qwen-code | 3180.0 | 758,211 | ||
| GPT-5.6 Sol Pro | xhigh, b, serial | codex | 3175.0 | $260 | 1,102,082 |
| Claude Fable 5 | xhigh, b, subagent | claude-code | 3164.0 | $237 | 363,518 |
| GPT-5.6 Sol | xhigh, b, serial | codex-goal | 3164.0 | $327 | 465,821 |
| GPT-5.6 Sol | xhigh, a, serial | codex-goal | 3160.0 | $419 | 522,005 |
| GLM-5.2 | high, a, subagent | pi | 3150.0 | $17 | 1,722,229 |
| Kimi K3 | max, b, serial | kimi-code-goal | 3146.0 | $66 | 237,149 |
| max, c, serial | claude-code | 3141.0 | 1,788,174 | ||
| GPT-5.6 Sol | xhigh, c, subagent | codex | 3136.0 | $436 | 435,060 |
| Grok 4.5 | xhigh, b, subagent | grok-cli | 3135.0 | $285 | 634,576 |
| GPT-5.6 Sol | xhigh, a, subagent | prime-agent | 3135.0 | $136 | 1,268,909 |
| Grok 4.5 | xhigh, a, subagent | grok-cli | 3132.0 | $34 | 186,468 |
| Grok 4.5 | xhigh, c, subagent | grok-cli | 3120.0 | $65 | 385,127 |
| max, b, serial | qwen-code | 3120.0 | 628,806 | ||
| Claude Sonnet 5 | max, b, subagent | claude-code | 3116.0 | $315 | 828,069 |
| Claude Fable 5 | low, b, subagent | claude-code | 3115.0 | $486 | 181,167 |
| Claude Fable 5 | medium, b, subagent | claude-code | 3115.0 | $253 | 209,645 |
| GPT-5.6 Sol | xhigh, b, subagent | prime-agent | 3113.0 | $157 | 1,182,991 |
| GPT-5.6 Sol | xhigh, c, serial | prime-agent | 3112.0 | $75 | 478,976 |
| GPT-5.6 Luna | xhigh, a, subagent | codex | 3110.0 | $21 | 887,636 |
| Claude Fable 5 | max, d, subagent | claude-code | 3107.0 | $617 | 604,488 |
| Claude Sonnet 5 | max, a, subagent | claude-code | 3105.0 | $462 | 2,104,701 |
| GPT-5.6 Sol | xhigh, a, serial | prime-agent | 3094.0 | $126 | 1,017,827 |
| Kimi K3 | max, b, subagent | claude-code | 3090.0 | $149 | 1,387,185 |
| Claude Fable 5 | max, a, subagent | claude-code | 3080.0 | $624 | 551,078 |
| Claude Fable 5 | low, a, subagent | claude-code | 3070.0 | $375 | 234,906 |
| Claude Fable 5 | low, b, subagent | claude-code | 3070.0 | $282 | 164,533 |
| Claude Fable 5 | medium, a, subagent | claude-code | 3070.0 | $331 | 309,826 |
| GPT-5.6 Sol Pro | xhigh, a, serial | codex | 3058.0 | $1,239 | 4,626,224 |
| Claude Fable 5 | low, a, subagent | claude-code | 3050.0 | $281 | 174,626 |
| Kimi K3 | max, b, serial | claude-code-goal | 3050.0 | $77 | 256,330 |
| Kimi K3 | max, a, serial | kimi-code-goal | 3048.0 | $277 | 596,474 |
| Kimi K3 | max, c, subagent | kimi-code | 3045.0 | $78 | 681,145 |
| GPT-5.6 Sol | xhigh, a, subagent | codex | 3042.0 | $1,631 | 2,232,409 |
| Kimi K3 | max, d, serial | claude-code | 3040.0 | $29 | 176,365 |
| Kimi K3 | max, a, subagent | claude-code | 3035.0 | $142 | 585,909 |
| Claude Fable 5 | high, a, subagent | claude-code | 3030.0 | $1,006 | 653,743 |
| Claude Fable 5 | high, b, subagent | claude-code | 3026.0 | $1,160 | 389,147 |
| Claude Opus 4.8 | max, a, subagent | claude-code | 3018.0 | $678 | 2,319,720 |
| Claude Fable 5 | xhigh, b, subagent | claude-code | 3010.0 | $697 | 563,336 |
| Claude Fable 5 | xhigh, a, serial | claude-code-goal | 3000.0 | $1,240 | 344,756 |
| Claude Fable 5 | xhigh, a, subagent | claude-code | 3000.0 | $400 | 404,404 |
| Claude Fable 5 | xhigh, b, subagent | prime-agent | 3000.0 | $1,331 | 1,903,937 |
| Claude Fable 5 | xhigh, b, subagent | claude-code | 2980.0 | $826 | 563,648 |
| Claude Opus 5 | high, a, serial | prime-agent | 2980.0 | $56 | 350,377 |
| Kimi K3 | max, a, subagent | kimi-code | 2974.0 | $337 | 1,413,952 |
| Claude Fable 5 | max, b, subagent | claude-code | 2970.0 | $765 | 670,572 |
| Claude Fable 5 | xhigh, a, subagent | claude-code | 2965.0 | $731 | 541,773 |
| Claude Fable 5 | xhigh, c, subagent | claude-code | 2938.0 | $2,184 | 778,970 |
| Claude Opus 5 | max, a, serial | claude-code | 2920.0 | $868 | 690,234 |
| Claude Fable 5 | xhigh, a, subagent | claude-code | 2915.0 | $676 | 580,489 |
| Claude Fable 5 | xhigh, c, subagent | prime-agent | 2900.0 | $1,036 | 1,550,547 |
| Claude Fable 5 | max, c, subagent | claude-code | 2850.0 | $6,663 | 2,208,803 |
| Claude Fable 5 | xhigh, a, subagent | prime-agent | 2836.0 | $1,116 | 1,128,993 |
| Claude Fable 5 | high, c, subagent | claude-code | 2726.0 | $3,201 | 1,105,606 |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official NanoGPT Speedrun Frontier leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| max, b, subagent | claude-code-sub | 3288.0 | 1,256,275 | ||
| xhigh, c, serial | claude-code | 3280.0 | $195 | 949,721 | |
| GPT-5.5 | xhigh, c, serial | codex | 3270.0 | $262 | 164,546 |
| max, a, subagent | prime-agent-sub | 3270.0 | 456,743 | ||
| Grok 4.5 | xhigh, a, serial | grok-cli-goal | 3270.0 | $112 | 110,871 |
| Grok 4.5 | xhigh, b, serial | grok-cli-goal | 3270.0 | $150 | 162,676 |
| xhigh, b, subagent | muse-code-sub | 3270.0 | 1,373,908 | ||
| max, a, serial | claude-code | 3265.0 | 331,253 | ||
| max, d, serial | claude-code | 3263.0 | 219,121 | ||
| max, a, subagent | claude-code-sub | 3263.0 | 703,172 | ||
| xhigh, c, serial | muse-code | 3263.0 | 2,362,863 | ||
| GLM-5.2 | xhigh, d, serial | claude-code | 3262.0 | $30 | 430,697 |
| Kimi K2.7 Code | max, b, subagent | kimi-code | 3260.0 | $34 | 727,053 |
| Kimi K2.7 Code | high, a, subagent | pi | 3260.0 | $20 | 294,328 |
| Claude Sonnet 5 | xhigh, b, subagent | claude-code | 3260.0 | $21 | 207,800 |
| xhigh, a, serial | claude-code-goal | 3260.0 | $97 | 675,416 | |
| Grok 4.6 | xhigh, b, subagent | grok-cli-sub | 3255.0 | 257,781 | |
| Kimi K2.7 Code | max, d, serial | kimi-code | 3251.0 | $6.01 | 1,501,977 |
| max, a, serial | prime-agent | 3250.0 | 1,820,857 | ||
| Claude Opus 4.8 | xhigh, b, subagent | claude-code | 3250.0 | $75 | 343,451 |
| xhigh, a, subagent | claude-code | 3250.0 | $324 | 1,343,151 | |
| Claude Opus 5 | max, b, serial | claude-code | 3250.0 | $51 | 210,171 |
| max, a, subagent | claude-code-sub | 3250.0 | 609,916 | ||
| GPT-5.6 Luna | xhigh, c, subagent | codex | 3247.0 | $21 | 859,634 |
| GPT-5.6 Terra | xhigh, c, serial | codex | 3245.0 | $84 | 219,454 |
| GPT-5.5 | xhigh, b, serial | codex | 3243.0 | $251 | 157,766 |
| GLM-5.2 | xhigh, a, serial | claude-code | 3243.0 | $27 | 385,518 |
| Grok 4.6 | xhigh, b, serial | grok-cli | 3243.0 | 287,723 | |
| GPT-5.6 Terra | xhigh, b, serial | codex | 3242.0 | $47 | 186,103 |
| Kimi K2.7 Code | max, a, subagent | kimi-code | 3240.0 | $51 | 763,161 |
| GLM-5.2 | xhigh, b, serial | claude-code | 3240.0 | $14 | 249,251 |
| max, b, subagent | pi | 3240.0 | $20 | 1,528,524 | |
| max, b, subagent | claude-code-sub | 3240.0 | 488,662 | ||
| xhigh, b, serial | muse-code | 3240.0 | 2,572,145 | ||
| Grok 4.6 | xhigh, c, serial | grok-cli | 3240.0 | 279,394 | |
| Claude Opus 4.8 | max, b, subagent | claude-code | 3236.0 | $366 | 1,466,322 |
| GPT-5.5 | xhigh, a, serial | codex | 3234.0 | $158 | 77,425 |
| max, a, subagent | pi | 3232.0 | $31 | 1,603,714 | |
| Claude Sonnet 5 | xhigh, a, subagent | claude-code | 3230.0 | $21 | 205,476 |
| Kimi K3 | max, a, serial | claude-code-goal | 3230.0 | $101 | 1,081,557 |
| xhigh, a, subagent | muse-code-sub | 3230.0 | 910,087 | ||
| Grok 4.6 | xhigh, a, serial | grok-cli | 3227.0 | 305,686 | |
| Claude Opus 4.8 | xhigh, a, subagent | claude-code | 3226.0 | $75 | 470,614 |
| max, b, serial | prime-agent | 3225.0 | 2,613,844 | ||
| Grok 4.6 | xhigh, a, subagent | grok-cli-sub | 3220.0 | 345,731 | |
| Kimi K3 | max, b, subagent | kimi-code | 3219.0 | $28 | 381,742 |
| GLM-5.2 | xhigh, c, subagent | claude-code | 3217.0 | $42 | 530,685 |
| GPT-5.6 Terra | xhigh, a, serial | codex | 3214.0 | $92 | 297,651 |
| max, c, serial | claude-code | 3205.0 | 319,065 | ||
| GLM-5.2 | xhigh, a, subagent | claude-code | 3205.0 | $30 | 536,371 |
| GLM-5.2 | xhigh, a, serial | claude-code-goal | 3201.0 | $68 | 548,989 |
| GLM-5.2 | xhigh, c, serial | claude-code | 3190.0 | $21 | 316,234 |
| GPT-5.6 Luna | xhigh, b, subagent | codex | 3185.0 | $15 | 730,250 |
| GPT-5.6 Sol | xhigh, b, subagent | codex | 3184.0 | $218 | 410,516 |
| max, c, serial | qwen-code | 3180.0 | 758,211 | ||
| GPT-5.6 Sol Pro | xhigh, b, serial | codex | 3175.0 | $260 | 1,102,082 |
| Claude Fable 5 | xhigh, b, subagent | claude-code | 3164.0 | $237 | 363,518 |
| GPT-5.6 Sol | xhigh, b, serial | codex-goal | 3164.0 | $327 | 465,821 |
| GPT-5.6 Sol | xhigh, a, serial | codex-goal | 3160.0 | $419 | 522,005 |
| GLM-5.2 | high, a, subagent | pi | 3150.0 | $17 | 1,722,229 |
| Kimi K3 | max, b, serial | kimi-code-goal | 3146.0 | $66 | 237,149 |
| max, c, serial | claude-code | 3141.0 | 1,788,174 | ||
| GPT-5.6 Sol | xhigh, c, subagent | codex | 3136.0 | $436 | 435,060 |
| Grok 4.5 | xhigh, b, subagent | grok-cli | 3135.0 | $285 | 634,576 |
| GPT-5.6 Sol | xhigh, a, subagent | prime-agent | 3135.0 | $136 | 1,268,909 |
| Grok 4.5 | xhigh, a, subagent | grok-cli | 3132.0 | $34 | 186,468 |
| Grok 4.5 | xhigh, c, subagent | grok-cli | 3120.0 | $65 | 385,127 |
| max, b, serial | qwen-code | 3120.0 | 628,806 | ||
| Claude Sonnet 5 | max, b, subagent | claude-code | 3116.0 | $315 | 828,069 |
| Claude Fable 5 | low, b, subagent | claude-code | 3115.0 | $486 | 181,167 |
| Claude Fable 5 | medium, b, subagent | claude-code | 3115.0 | $253 | 209,645 |
| GPT-5.6 Sol | xhigh, b, subagent | prime-agent | 3113.0 | $157 | 1,182,991 |
| GPT-5.6 Sol | xhigh, c, serial | prime-agent | 3112.0 | $75 | 478,976 |
| GPT-5.6 Luna | xhigh, a, subagent | codex | 3110.0 | $21 | 887,636 |
| Claude Fable 5 | max, d, subagent | claude-code | 3107.0 | $617 | 604,488 |
| Claude Sonnet 5 | max, a, subagent | claude-code | 3105.0 | $462 | 2,104,701 |
| GPT-5.6 Sol | xhigh, a, serial | prime-agent | 3094.0 | $126 | 1,017,827 |
| Kimi K3 | max, b, subagent | claude-code | 3090.0 | $149 | 1,387,185 |
| Claude Fable 5 | max, a, subagent | claude-code | 3080.0 | $624 | 551,078 |
| Claude Fable 5 | low, a, subagent | claude-code | 3070.0 | $375 | 234,906 |
| Claude Fable 5 | low, b, subagent | claude-code | 3070.0 | $282 | 164,533 |
| Claude Fable 5 | medium, a, subagent | claude-code | 3070.0 | $331 | 309,826 |
| GPT-5.6 Sol Pro | xhigh, a, serial | codex | 3058.0 | $1,239 | 4,626,224 |
| Claude Fable 5 | low, a, subagent | claude-code | 3050.0 | $281 | 174,626 |
| Kimi K3 | max, b, serial | claude-code-goal | 3050.0 | $77 | 256,330 |
| Kimi K3 | max, a, serial | kimi-code-goal | 3048.0 | $277 | 596,474 |
| Kimi K3 | max, c, subagent | kimi-code | 3045.0 | $78 | 681,145 |
| GPT-5.6 Sol | xhigh, a, subagent | codex | 3042.0 | $1,631 | 2,232,409 |
| Kimi K3 | max, d, serial | claude-code | 3040.0 | $29 | 176,365 |
| Kimi K3 | max, a, subagent | claude-code | 3035.0 | $142 | 585,909 |
| Claude Fable 5 | high, a, subagent | claude-code | 3030.0 | $1,006 | 653,743 |
| Claude Fable 5 | high, b, subagent | claude-code | 3026.0 | $1,160 | 389,147 |
| Claude Opus 4.8 | max, a, subagent | claude-code | 3018.0 | $678 | 2,319,720 |
| Claude Fable 5 | xhigh, b, subagent | claude-code | 3010.0 | $697 | 563,336 |
| Claude Fable 5 | xhigh, a, serial | claude-code-goal | 3000.0 | $1,240 | 344,756 |
| Claude Fable 5 | xhigh, a, subagent | claude-code | 3000.0 | $400 | 404,404 |
| Claude Fable 5 | xhigh, b, subagent | prime-agent | 3000.0 | $1,331 | 1,903,937 |
| Claude Fable 5 | xhigh, b, subagent | claude-code | 2980.0 | $826 | 563,648 |
| Claude Opus 5 | high, a, serial | prime-agent | 2980.0 | $56 | 350,377 |
| Kimi K3 | max, a, subagent | kimi-code | 2974.0 | $337 | 1,413,952 |
| Claude Fable 5 | max, b, subagent | claude-code | 2970.0 | $765 | 670,572 |
| Claude Fable 5 | xhigh, a, subagent | claude-code | 2965.0 | $731 | 541,773 |
| Claude Fable 5 | xhigh, c, subagent | claude-code | 2938.0 | $2,184 | 778,970 |
| Claude Opus 5 | max, a, serial | claude-code | 2920.0 | $868 | 690,234 |
| Claude Fable 5 | xhigh, a, subagent | claude-code | 2915.0 | $676 | 580,489 |
| Claude Fable 5 | xhigh, c, subagent | prime-agent | 2900.0 | $1,036 | 1,550,547 |
| Claude Fable 5 | max, c, subagent | claude-code | 2850.0 | $6,663 | 2,208,803 |
| Claude Fable 5 | xhigh, a, subagent | prime-agent | 2836.0 | $1,116 | 1,128,993 |
| Claude Fable 5 | high, c, subagent | claude-code | 2726.0 | $3,201 | 1,105,606 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official NanoGPT Speedrun Frontier leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official NanoGPT Speedrun Frontier leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in AI research
- PaperBench In use Agent replicates ICML 2024 papers from scratch. LLM judge grades against author-written hierarchical rubrics.
Qwen 3.8 Max 93%
- MLS-Bench In use Agent must invent one algorithmic change in an ML research scaffold whose gain transfers across settings. Scored relative to human baselines. Claude Fable 5.1 50.3%
- PostTrainBench 1.2 In use CLI agent gets a base model, one H100 and a time budget to post-train it for a target benchmark. Scored on normalized gain. Claude Fable 5.1 44.6%
- MLE-bench In use Kaggle-style ML engineering competitions. Agent trains models and is scored against competition leaderboards. GPT-6.2 Astra 96.1%
- PMPP-Hard In use Agents write GPU kernels for 69 tasks that must be correct, and on 48 of them also run within 1.25 times the reference time. The score is the area under the solved-tasks curve over a 10-hour budget, out of 690 task-hours. Kimi K3 0.71
- KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup.
MiniMax-M3 28.8%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark
Catalog record 1 warning 1 pass
- Warning Task access We don’t know whether the tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (19 entries).
- Pass Models scored 18 models we track, plus 1 other entry on its leaderboard.
- Pass Newest score The newest is from August 2026, for Grok 4.6.
- Note Frontier models None of the top 10 has a score. The strongest tested is Claude Opus 5 (#11).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on primeintellect.ai: 109 rows for models we track and 1 other entry at the last read.
- Pass Last read November 21st, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.