Benchmark Almanac

METR Time Horizon 2

GPT-6 Astra 27.1 h leads since November 2026
Not in model cards
Version 2

Agents complete 36 tasks across 12 families. METR estimates how long a skilled human would need for tasks that agents complete with 50 percent reliability.

METR typically ran 15 models. The published comparison adds short tasks from version 1.1 to the verifiable subset of TH2.

Made by
METR
Released
November 2026
Kind
Task length
Level
Research frontier
Scored
Checked automatically
Answers
Agent task
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text → Actions
Interaction
Agentic
Run by
A host
In views
Agents
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
September 2026
Scores refreshed
November 21st, 2026
11
Models scored on its own leaderboard

GPT-6 Astra leads with 27.1 h among models released by September 2026

Best score per model from the benchmark's own leaderboard, read on November 21st, 2026. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-6 Astra TH2 (Verifiable) 27.1 h
  2. 2 Claude Opus 5.5 TH2 (Verifiable) 19.9 h
  3. 3 Claude Fable 5.1 TH2 (Verifiable) 14.9 h
  4. 4 Claude Opus 5 TH2 (Verifiable) 10.1 h
  5. 5 GPT-5.5 TH2 (Verifiable) 6.2 h
  6. 6 Claude Fable 5 TH2 (Verifiable) 6.1 h
  7. 7 Claude Opus 4.8 TH2 (Verifiable) 3.6 h
  8. 8 Claude Opus 4.7 TH2 (Verifiable) 2.5 h
  9. 9 GPT-5.6 Sol TH2 (Verifiable) 2.4 h
  10. 10 GPT-5.4 TH2 (Verifiable) 83 min
  11. 11 Claude Opus 4.6 TH2 (Verifiable) 64 min

Source: Official METR Time Horizon 2 leaderboard

#ModelScore ±SettingReleasedRun onSource
1 GPT-6 Astra 27.1 h 438.8 to 28353 TH2 (Verifiable) Sep 2026 Sep 2026 Official leaderboard
2 Claude Opus 5.5 19.9 h 327.8 to 33883 TH2 (Verifiable) Sep 2026 Sep 2026 Official leaderboard
3 Claude Fable 5.1 14.9 h 319.9 to 12634 TH2 (Verifiable) Sep 2026 Sep 2026 Official leaderboard
4 Claude Opus 5 10.1 h 242.9 to 2845.2 TH2 (Verifiable) Jul 2026 Jul 2026 Official leaderboard
5 GPT-5.5 6.2 h 93.9 to 930.8 TH2 (Verifiable) Apr 2026 Apr 2026 Official leaderboard
6 Claude Fable 5 6.1 h 131.8 to 1174.4 TH2 (Verifiable) Jun 2026 Jun 2026 Official leaderboard
7 Claude Opus 4.8 3.6 h 60.8 to 618.8 TH2 (Verifiable) May 2026 May 2026 Official leaderboard
8 Claude Opus 4.7 2.5 h 56.3 to 574.2 TH2 (Verifiable) Apr 2026 Apr 2026 Official leaderboard
9 GPT-5.6 Sol 2.4 h 71.6 to 1075.5 TH2 (Verifiable) Jul 2026 Jul 2026 Official leaderboard
10 GPT-5.4 83 min 31.9 to 421.1 TH2 (Verifiable) Mar 2026 Mar 2026 Official leaderboard
11 Claude Opus 4.6 64 min 32.4 to 388.2 TH2 (Verifiable) Feb 2026 Feb 2026 Official leaderboard

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard 1 leader
GPT-6 Astra 27.1 h

Source: Official METR Time Horizon 2 leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard
10 min 100 min 16.7 h 166.7 h 2026 10 min 100 min 16.7 h 166.7 h 2026

Source: Official METR Time Horizon 2 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.

Versions

More in AI research

All 164 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 Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (11 entries).
  • Pass Models scored 11 models we track.
  • Pass Newest score The newest is from September 2026, for Claude Opus 5.5.
  • Pass Frontier models 3 of the top 10 have a score, led by GPT-6 Astra (#4).
  • Note Headroom Scored as minutes, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on metr.org: 11 rows for models we track 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.

How we check