Benchmark Almanac

SAGE v1.1 (Vals)

Claude Opus 5.5 91.3% leads since Oct 2026 Not in model cards

Models grade handwritten undergraduate math and computer science work against rubrics. Scored by class-balanced accuracy of each rubric check against expert grading.

Vals changed the metric from v1, which moved strict graders up and lenient ones down. 545 tests in five suites. Claude Opus 5.5 led 28 models at 91.30% on 2026-10-27.

Made by Vals AI Released Oct 2026 Other Tasks kept private Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
University Undergraduates and entry-level professionals
Scored
Graded by people Human experts grade or compare the outputs.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
Leaderboard
28
Models scored on its own leaderboard

Claude Opus 5.5 leads with 91.3%

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.

28 models. Scroll sideways for the rest.

  1. 91.3 Claude Opus 5.5
  2. 90.5 GPT-6.1 Sol
  3. 89.8 Claude Haiku 5.5
  4. 89.5 Gemini 4 Argon
  5. 89.3 GLM-5.3-Flash
  6. 89.2 Qwen3.7-Plus
  7. 88.7 Grok 4.7
  8. 88.2 Gemini 3.8 Flash
  9. 88.0 Claude Sonnet 5.5
  10. 87.0 GPT-5.6 Terra
  11. 86.8 Kimi K3
  12. 86.4 Ember-1
  13. 85.8 Muse Spark Open
  14. 85.5 Qwen3.5 397B-A17B
  15. 85.5 MiniMax-M3
  16. 85.4 DeepSeek V4.1 Flash
  17. 85.3 MiniMax-M3.1-Flash
  18. 84.9 Qwen 3.5 Flash (hosted 35B-A3B)
  19. 84.5 MiMo-V2.6 Pro
  20. 83.4 GPT-6 Luna
  21. 83.3 MiMo-V2.6 Flash
  22. 83.2 Gemini 3.1 Pro
  23. 82.8 Gemma 4 31B IT
  24. 81.8 Qwen 3.8 Max
  25. 79.5 Gemini 3.5 Flash-Lite
  26. 74.5 Grok 4.1 Fast
  27. 72.0 Mistral Medium 3.5
  28. 59.2 Mistral Small 4
  1. 1 Claude Opus 5.5 91.3%
  2. 2 GPT-6.1 Sol 90.5%
  3. 3 Claude Haiku 5.5 89.8%
  4. 4 Gemini 4 Argon 89.5%
  5. 5 GLM-5.3-Flash 89.3%
  6. 6 Qwen3.7-Plus 89.2%
  7. 7 Grok 4.7 88.7%
  8. 8 Gemini 3.8 Flash 88.2%
  9. 9 Claude Sonnet 5.5 88%
  10. 10 GPT-5.6 Terra 87%
  11. 11 Kimi K3 86.8%
  12. 12 Ember-1 86.4%
  13. 13 Muse Spark Open 85.8%
  14. 14 Qwen3.5 397B-A17B 85.5%
  15. 15 MiniMax-M3 85.5%
  16. 16 DeepSeek V4.1 Flash 85.4%
  17. 17 MiniMax-M3.1-Flash 85.3%
  18. 18 Qwen 3.5 Flash (hosted 35B-A3B) 84.9%
  19. 19 MiMo-V2.6 Pro 84.5%
  20. 20 GPT-6 Luna 83.4%
  21. 21 MiMo-V2.6 Flash 83.3%
  22. 22 Gemini 3.1 Pro 83.2%
  23. 23 Gemma 4 31B IT 82.8%
  24. 24 Qwen 3.8 Max 81.8%
  25. 25 Gemini 3.5 Flash-Lite 79.5%
  26. 26 Grok 4.1 Fast 74.5%
  27. 27 Mistral Medium 3.5 72%
  28. 28 Mistral Small 4 59.2%
#ModelScoreSettingDateSource
1 Claude Opus 5.5 91.3% Sep 2026 Official leaderboard
2 GPT-6.1 Sol 90.5% Sep 2026 Official leaderboard
3 Claude Haiku 5.5 89.8% Oct 2026 Official leaderboard
4 Gemini 4 Argon 89.5% Sep 2026 Official leaderboard
5 GLM-5.3-Flash 89.3% Aug 2026 Official leaderboard
6 Qwen3.7-Plus 89.2% Jun 2026 Official leaderboard
7 Grok 4.7 88.7% Sep 2026 Official leaderboard
8 Gemini 3.8 Flash 88.2% Sep 2026 Official leaderboard
9 Claude Sonnet 5.5 88% Sep 2026 Official leaderboard
10 GPT-5.6 Terra 87% Jul 2026 Official leaderboard
11 Kimi K3 86.8% Jul 2026 Official leaderboard
12 Ember-1 86.4% Sep 2026 Official leaderboard
13 Muse Spark Open 85.8% Oct 2026 Official leaderboard
14 Qwen3.5 397B-A17B 85.5% Feb 2026 Official leaderboard
15 MiniMax-M3 85.5% Jun 2026 Official leaderboard
16 DeepSeek V4.1 Flash 85.4% Sep 2026 Official leaderboard
17 MiniMax-M3.1-Flash 85.3% Oct 2026 Official leaderboard
18 Qwen 3.5 Flash (hosted 35B-A3B) 84.9% Feb 2026 Official leaderboard
19 MiMo-V2.6 Pro 84.5% Sep 2026 Official leaderboard
20 GPT-6 Luna 83.4% Sep 2026 Official leaderboard
21 MiMo-V2.6 Flash 83.3% Sep 2026 Official leaderboard
22 Gemini 3.1 Pro 83.2% Feb 2026 Official leaderboard
23 Gemma 4 31B IT 82.8% Apr 2026 Official leaderboard
24 Qwen 3.8 Max 81.8% Aug 2026 Official leaderboard
25 Gemini 3.5 Flash-Lite 79.5% Jul 2026 Official leaderboard
26 Grok 4.1 Fast 74.5% Nov 2025 Official leaderboard
27 Mistral Medium 3.5 72% Apr 2026 Official leaderboard
28 Mistral Small 4 59.2% Mar 2026 Official leaderboard

Number one over time

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

Anthropic
Own leaderboard Passed 85% in Oct 2026, by Claude Opus 5.5
Claude Opus 5.5 91.3%

The best score passed 85% in Feb 2026

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

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Qwen3.7-Plus Qwen3.5 397B-A17B Grok 4.1 Fast 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Grok 4.1 Fast

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Google DeepMind Anthropic
SAGE v1.1 (Vals) Passed 85% in Oct 2026, by Claude Opus 5.5
Claude Opus 5.5 91.3%
SAGE (Vals) 5 leaders
Claude Opus 4.7 56.1%
SAGE v1.1 (Vals) Passed 85% in Oct 2026, by Claude Opus 5.5
Claude Opus 5.5 91.3%
Replaces SAGE (Vals)

More in Other

All 41 in Other
  • Personal Superintelligence Index (Meta) In use Meta's composite of seven personal and knowledge-work evals: Agents' Last Exam, AutomationBench-AA, FocusBench, JobBench, GDP.pdf, VBT and DRACO.
  • Artificial Analysis Intelligence Index v5 In use Version 5 of Artificial Analysis's index: nine evaluations in four equal categories, most with private test sets, run three times each in its own agent harness. Gemini 4 Carbon 59.0
  • Anthropic ECI (AECI) Only in system cards Index that combines a basket of capability benchmarks into one capability score to track trends across Anthropic models. Claude Opus 5.5 169.4
  • Qwen in-house English automatic evaluation Only in system cards Models answer Qwen in-house English datasets on knowledge, comprehension, coding, math and related skills. Skills and weights change between reports. Qwen2-72B 72.9%
  • Recognition of model-written user turns (Anthropic) Only in system cards Models write synthetic user turns for real Claude Code transcripts, then judge which version is real. Score is accuracy above chance. Claude Mythos Preview 0.35
  • LiveBench Dropped Models answer monthly refreshed questions across math, coding, reasoning, language, instruction following, and data analysis. Scored automatically against objective ground truth. Claude Fable 5.1 85.7%