FrontierML
Agents spend a fixed budget on GPUs and model APIs to train models for 21 tasks. An offline evaluator grades their submissions on held-out data.
On November 19, 2026, Claude Opus 5.5 led at 58.8, ahead of Claude Sonnet 5.5 at 50.4 and GPT-6 Astra at 44.5.
- Made by
- Induction Labs
- Released
- November 2026
- Kind
- Machine learning engineering
- Level
- Research frontier
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal, APIs and tools
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- October 2026
- Scores refreshed
- November 20th, 2026
Claude Opus 5.5 leads with 58.9
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
- 58.9 Claude Opus 5.5
- 50.5 Claude Sonnet 5.5
- 44.5 GPT-6 Astra
- 41.9 Claude Fable 5.1
- 39.5
Gemini 4 Argon
- 38.7 GPT-6.1 Sol
- 34.9 Claude Haiku 5.5
- 29.6 Kimi K3.1 Pro
- 24.0
MiniMax-M3.1-Flash
- 23.5
DeepSeek V4.1 Flash 1017
- 21.6
Muse Wonder
- 20.3 Kimi K3
- 19.9
Qwen 3.8 Max
- 19.6 GLM-5.3
- 16.0
Muse Spark 1.3
- 13.3 GPT-6 Luna
- 12.3 Grok 4.7
- 9.49
Gemini 3.8 Flash
- 9.29 Beam 1.0
- 1 Claude Opus 5.5 $200 wallet · xhigh 58.9
- 2 Claude Sonnet 5.5 $200 wallet · xhigh 50.5
- 3 GPT-6 Astra $200 wallet · xhigh 44.5
- 4 Claude Fable 5.1 $200 wallet · xhigh 41.9
- 5
Gemini 4 Argon $200 wallet · high 39.5
- 6 GPT-6.1 Sol $200 wallet · xhigh 38.7
- 7 Claude Haiku 5.5 $200 wallet · max 34.9
- 8 Kimi K3.1 Pro $200 wallet · max 29.6
- 9
MiniMax-M3.1-Flash $50 wallet · max 24.0
- 10
DeepSeek V4.1 Flash 1017 $200 wallet · max 23.5
- 11
Muse Wonder $10 wallet · max 21.6
- 12 Kimi K3 $200 wallet · max 20.3
- 13
Qwen 3.8 Max $10 wallet · xhigh 19.9
- 14 GLM-5.3 $200 wallet · max 19.6
- 15
Muse Spark 1.3 $200 wallet · max 16.0
- 16 GPT-6 Luna $50 wallet · xhigh 13.3
- 17 Grok 4.7 $10 wallet · xhigh 12.3
- 18
Gemini 3.8 Flash $50 wallet · high 9.49
- 19 Beam 1.0 $50 wallet · max 9.29
Source: Official FrontierML leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 58.9 | $200 wallet · xhigh, Claude Code | Sep 2026 | Official leaderboard |
| 2 | Claude Sonnet 5.5 | 50.5 | $200 wallet · xhigh, Claude Code | Sep 2026 | Official leaderboard |
| 3 | GPT-6 Astra | 44.5 | $200 wallet · xhigh, Codex CLI | Sep 2026 | Official leaderboard |
| 4 | Claude Fable 5.1 | 41.9 | $200 wallet · xhigh, Claude Code | Sep 2026 | Official leaderboard |
| 5 | | 39.5 | $200 wallet · high, Antigravity CLI | Sep 2026 | Official leaderboard |
| 6 | GPT-6.1 Sol | 38.7 | $200 wallet · xhigh, Codex CLI | Sep 2026 | Official leaderboard |
| 7 | Claude Haiku 5.5 | 34.9 | $200 wallet · max, Claude Code | Oct 2026 | Official leaderboard |
| 8 | Kimi K3.1 Pro | 29.6 | $200 wallet · max, Claude Code | Oct 2026 | Official leaderboard |
| 9 | | 24.0 | $50 wallet · max, Claude Code | Oct 2026 | Official leaderboard |
| 10 | | 23.5 | $200 wallet · max, Claude Code | Oct 2026 | Official leaderboard |
| 11 | | 21.6 | $10 wallet · max, Muse Code | Oct 2026 | Official leaderboard |
| 12 | Kimi K3 | 20.3 | $200 wallet · max, Kimi Code | Jul 2026 | Official leaderboard |
| 13 | | 19.9 | $10 wallet · xhigh, Claude Code | Aug 2026 | Official leaderboard |
| 14 | GLM-5.3 | 19.6 | $200 wallet · max, Claude Code | Aug 2026 | Official leaderboard |
| 15 | | 16.0 | $200 wallet · max, Muse Code | Sep 2026 | Official leaderboard |
| 16 | GPT-6 Luna | 13.3 | $50 wallet · xhigh, Codex CLI | Sep 2026 | Official leaderboard |
| 17 | Grok 4.7 | 12.3 | $10 wallet · xhigh, Grok Build | Sep 2026 | Official leaderboard |
| 18 | | 9.49 | $50 wallet · high, Antigravity CLI | Sep 2026 | Official leaderboard |
| 19 | Beam 1.0 | 9.29 | $50 wallet · max, Mirror CLI | Oct 2026 | Official leaderboard |
The best score, 58.9, costs $183 a task
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official FrontierML leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 5.5 | $200 wallet · xhigh | Claude Code | 58.9 | $183 | 295,997 |
| Claude Opus 5.5 | $50 wallet · xhigh | Claude Code | 52.4 | $87 | 295,932 |
| Claude Sonnet 5.5 | $200 wallet · xhigh | Claude Code | 50.5 | $172 | 754,369 |
| Claude Sonnet 5.5 | $50 wallet · xhigh | Claude Code | 47.3 | $72 | 555,161 |
| Claude Sonnet 5.5 | $10 wallet · xhigh | Claude Code | 45.5 | $37 | 440,322 |
| Claude Opus 5.5 | $10 wallet · xhigh | Claude Code | 45.4 | $51 | 274,396 |
| GPT-6 Astra | $200 wallet · xhigh | Codex CLI | 44.5 | $217 | 524,342 |
| GPT-6 Astra | $50 wallet · xhigh | Codex CLI | 42.8 | $157 | 539,380 |
| Claude Fable 5.1 | $200 wallet · xhigh | Claude Code | 41.9 | $182 | 251,043 |
| $200 wallet · high | Antigravity CLI | 39.5 | $67 | 489,467 | |
| GPT-6.1 Sol | $200 wallet · xhigh | Codex CLI | 38.7 | $114 | 393,167 |
| GPT-6 Astra | $10 wallet · xhigh | Codex CLI | 37.6 | $85 | 350,254 |
| Claude Fable 5.1 | $50 wallet · xhigh | Claude Code | 37.4 | $115 | 231,254 |
| GPT-6.1 Sol | $50 wallet · xhigh | Codex CLI | 36.2 | $38 | 357,351 |
| Claude Fable 5.1 | $10 wallet · xhigh | Claude Code | 35.8 | $74 | 228,225 |
| Claude Haiku 5.5 | $200 wallet · max | Claude Code | 34.9 | $69 | 449,815 |
| GPT-6.1 Sol | $10 wallet · xhigh | Codex CLI | 31.6 | $14 | 247,939 |
| $50 wallet · high | Antigravity CLI | 30.4 | $41 | 555,767 | |
| Kimi K3.1 Pro | $200 wallet · max | Claude Code | 29.6 | $71 | 340,582 |
| $10 wallet · high | Antigravity CLI | 29.6 | $25 | 615,187 | |
| Claude Haiku 5.5 | $50 wallet · max | Claude Code | 29.1 | $34 | 427,452 |
| Claude Haiku 5.5 | $10 wallet · max | Claude Code | 28.2 | $13 | 379,477 |
| Kimi K3.1 Pro | $50 wallet · max | Claude Code | 26.7 | $43 | 259,663 |
| Kimi K3.1 Pro | $10 wallet · max | Claude Code | 25.5 | $18 | 222,013 |
| $50 wallet · max | Claude Code | 24.0 | $34 | 1,097,404 | |
| $200 wallet · max | Claude Code | 23.5 | $37 | 419,695 | |
| $200 wallet · max | Claude Code | 23.2 | $42 | 928,317 | |
| $50 wallet · max | Claude Code | 22.5 | $20 | 442,954 | |
| $10 wallet · max | Muse Code | 21.6 | $23 | 189,607 | |
| $10 wallet · max | Claude Code | 21.3 | $6.09 | 398,478 | |
| Kimi K3 | $200 wallet · max | Kimi Code | 20.3 | $110 | 257,525 |
| $10 wallet · xhigh | Claude Code | 19.9 | $20 | 297,216 | |
| GLM-5.3 | $200 wallet · max | Claude Code | 19.6 | $53 | 411,466 |
| $50 wallet · xhigh | Claude Code | 19.4 | $43 | 373,638 | |
| $200 wallet · max | Muse Code | 19.3 | $50 | 183,353 | |
| $10 wallet · max | Claude Code | 19.2 | $9.14 | 800,166 | |
| GLM-5.3 | $10 wallet · max | Claude Code | 18.6 | $16 | 381,035 |
| Kimi K3 | $50 wallet · max | Kimi Code | 17.8 | $72 | 248,428 |
| $200 wallet · xhigh | Claude Code | 17.5 | $83 | 427,035 | |
| GLM-5.3 | $50 wallet · max | Claude Code | 17.4 | $38 | 432,322 |
| $50 wallet · max | Muse Code | 17.2 | $36 | 196,746 | |
| Kimi K3 | $10 wallet · max | Kimi Code | 16.2 | $50 | 274,684 |
| $200 wallet · max | Muse Code | 16.0 | $59 | 361,425 | |
| $10 wallet · max | Muse Code | 14.1 | $21 | 366,217 | |
| GPT-6 Luna | $50 wallet · xhigh | Codex CLI | 13.3 | $12 | 217,520 |
| GPT-6 Luna | $10 wallet · xhigh | Codex CLI | 13.2 | $4.77 | 184,658 |
| GPT-6 Luna | $200 wallet · xhigh | Codex CLI | 13.1 | $18 | 211,329 |
| $50 wallet · max | Muse Code | 12.6 | $35 | 337,600 | |
| Grok 4.7 | $10 wallet · xhigh | Grok Build | 12.3 | $33 | 122,650 |
| Grok 4.7 | $200 wallet · xhigh | Grok Build | 10.9 | $50 | 117,424 |
| Grok 4.7 | $50 wallet · xhigh | Grok Build | 10.8 | $43 | 121,743 |
| $50 wallet · high | Antigravity CLI | 9.49 | $17 | 416,300 | |
| Beam 1.0 | $50 wallet · max | Mirror CLI | 9.29 | $23 | 71,508 |
| $10 wallet · high | Antigravity CLI | 8.13 | $12 | 378,017 | |
| Beam 1.0 | $10 wallet · max | Mirror CLI | 7.49 | $9.92 | 60,469 |
| $200 wallet · high | Antigravity CLI | 5.70 | $13 | 238,425 | |
| Beam 1.0 | $200 wallet · max | Mirror CLI | 3.95 | $33 | 77,398 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official FrontierML 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 FrontierML 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
All 10 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (19 entries).
- Pass Models scored 19 models we track.
- Pass Newest score The newest is from October 2026, for Kimi K3.1 Pro.
- Pass Frontier models 6 of the top 10 have a score, led by GPT-6 Astra (#4).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on frontierml.org: 57 rows for models we track at the last read.
- Pass Last read November 20th, 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.