Benchmark Almanac

Internal AI research evaluation suite 2

Claude Mythos Preview 0.65 leads since April 2026
Only in system cards

Agents perform scaled-down ML and alignment research tasks similar to past Anthropic research projects.

In-house Anthropic evaluation. The tasks are not public.

Made by
Anthropic
Released
May 2025
Kind
AI research tasks
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Opus 4.6 #37 on the capability index
Last in a system card
April 2026
Last model added
April 2026
6
Models scored in lab cards
6
System cards report it, from 1 lab

Claude Mythos Preview leads with 0.65 among models released by April 2026

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

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

  1. 0.65 Claude Mythos Preview
  2. 0.61 Claude Opus 4.6
  3. 0.60 Claude Opus 4.5
  4. 0.51 Claude Sonnet 4.5
  5. 0.36 Claude Sonnet 4
  6. 0.35 Claude Opus 4
  1. 1 Claude Mythos Preview 0.65
  2. 2 Claude Opus 4.6 0.61
  3. 3 Claude Opus 4.5 0.60
  4. 4 Claude Sonnet 4.5 0.51
  5. 5 Claude Sonnet 4 0.36
  6. 6 Claude Opus 4 0.35

Source: Labs’ model cards

#ModelScoreSettingDateSource
1 Claude Mythos Preview 0.65 Apr 2026 Lab model card
2 Claude Opus 4.6 0.61 Feb 2026 Lab model card
3 Claude Opus 4.5 0.60 Nov 2025 Lab model card
4 Claude Sonnet 4.5 0.51 Sep 2025 Lab model card
5 Claude Sonnet 4 0.36 May 2025 Lab model card
6 Claude Opus 4 0.35 May 2025 Lab model card

5 models have held number one in labs’ own numbers since its release in May 2025

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

Anthropic
Lab claims Not tested since Apr 2026
Claude Mythos Preview 0.65

Source: Labs’ model cards

Scores over time

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

Reported by the lab in its model card
0.40 0.50 0.60 0.70 2025 2026 0.40 0.50 0.60 0.70 2025 2026

Source: Labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

6 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.

DateDocumentScoreAs the document writes itSection
Apr 7, 2026 Claude Mythos Preview system card 0.65 Internal suite 2, score 0.65, rule-out threshold 0.6 (Table 2.3.3.A) AI R&D risk
Feb 5, 2026 Claude Opus 4.6 system card 0.612 Internal AI research evaluation suite 2, weighted average score 0.6124, rule-out threshold 0.6 (8.3.4) AI R&D risk
Nov 24, 2025 Claude Opus 4.5 system card 0.604 Internal AI research evaluation suite 2, weighted average score 0.604, rule-out threshold 0.6 (7.3.3) AI R&D risk
Sep 29, 2025 Claude Sonnet 4.5 system card 0.514 Internal AI research evaluation suite 2, weighted average score 0.514, rule-out threshold 0.6 (9.3.4) AI R&D risk
May 22, 2025 Claude Sonnet 4 system card 0.365 Internal AI research evaluation suite 2, weighted average score 0.365 (7.3.4) AI R&D risk
May 22, 2025 Claude Opus 4 system card 0.355 Internal AI research evaluation suite 2, weighted average score 0.355 (7.3.4) AI R&D risk

More in AI research

All 104 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 Astra 93.8%
  • KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup. MiniMax-M3 28.8%
  • 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. Inkling 2 Small 494.3