NanoGPT Speedrun
Improve a small GPT training script to reach a target validation loss as fast as possible on fixed hardware.
OpenAI and METR run their own constrained variants on one H100 or a budget. METR used it in July 2026 to define an expenditure horizon, the budget where humans become cheaper than agents.
- Made by
- Keller Jordan (community), OpenAI and METR (variants)
- Released
- June 2024
- Announced on X
- 129K views, 992 likes
- Kind
- Machine learning engineering
- Level
- Research frontier
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Efficiency
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- GPT-6.2 Astra #1 on the capability index
- Last in a system card
- November 2026
- Last model added
- November 2026
GPT-6.2 Astra leads with 42.9%
Best score per model as labs report them in their own cards, drawn striped. Labs pick their own settings, so these compare less cleanly than independent runs.
- 1 GPT-6.2 Astra 42.9%
- 2 GPT-6.2 Sol 14.6%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6.2 Astra | 42.9% | Nov 2026 | Lab model card | |
| 2 | GPT-6.2 Sol | 14.6% | Nov 2026 | Lab model card |
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
Labs report it only in system cards
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark. Some system cards do, below.
9 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests. OpenAI accounts for 6 of the 6 releases that named it since 2024.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 17, 2026 | GPT-6.2 Sol system card | 14.6 | NanoGPT, mean share of baseline training time saved, version 5 (Table 5) | AI R&D risk |
| Nov 17, 2026 | GPT-6.2 Astra system card | 42.9 | NanoGPT, mean share of baseline training time saved, version 5 (Table 5) | AI R&D risk |
| Sep 3, 2026 | GPT-6 Astra system card | NanoGPT, mean normalized reward vs output tokens, chart without printed numbers (Figure 54) | AI R&D risk | |
| Jul 9, 2026 | GPT-5.6 Terra system card | NanoGPT, mean reward by cost and latency, chart without printed numbers (Figures 41 to 44, 9.1.3.3) | AI R&D risk | |
| Jul 9, 2026 | GPT-5.6 Sol system card | NanoGPT, mean reward by cost and latency, chart without printed numbers (Figures 41 to 44, 9.1.3.3) | AI R&D risk | |
| Jul 9, 2026 | GPT-5.6 Luna system card | NanoGPT, mean reward by cost and latency, chart without printed numbers (Figures 41 to 44, 9.1.3.3) | AI R&D risk | |
| Jun 25, 2026 | GPT-5.6 Terra system card | NanoGPT, mean reward by cost and latency, chart without printed numbers (Figures 41 to 44, 9.1.3.3) | AI R&D risk | |
| Jun 25, 2026 | GPT-5.6 Sol system card | NanoGPT, mean reward by cost and latency, chart without printed numbers (Figures 41 to 44, 9.1.3.3) | AI R&D risk | |
| Jun 25, 2026 | GPT-5.6 Luna system card | NanoGPT, mean reward by cost and latency, chart without printed numbers (Figures 41 to 44, 9.1.3.3) | AI R&D risk |
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
All 5 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Note Independent runs Only labs’ own numbers, so no one else has checked them. Scores from system cards (2 scores).
- Note Models scored Only 2 models we track.
- Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
- Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6.2 Astra (#1).
- Pass Headroom The best is 42.9%, under 85%, so it still separates models.
Leaderboard import 1 note
- Note Importer No public leaderboard to read.
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.