BenchLeader
Menu
Best LLMsChoose a modelCoding plansLeaderboardsModelsBenchmarksProvidersReleases7

CritPt leaderboard

As of 6 Nov 2026, the top score on CritPt is 32.9% by Muse Wonder Full (Meta), ahead of GPT-5.6 Sol at 32.3%, across 568 model configurations with a published result (Artificial Analysis, result dated 12 Oct 2026). CritPt asks models unpublished physics research problems at the level of a first PhD project, written by researchers who know the answers. CritPt is shown for reference only and does not count toward the BenchLeader Index.

Key facts (as of 6 Nov 2026)

  • Muse Wonder Full holds the top CritPt score at 32.9% (Artificial Analysis, 12 Oct 2026).
  • GPT-5.6 Sol is second on CritPt at 32.3% (Artificial Analysis, 9 Jul 2026).
  • CritPt: Unpublished physics research problems at the level of a first PhD project. Run by Artificial Analysis. Index v5's hard subset replaced it in the BenchLeader Index (Artificial Analysis).
  • 568 model configurations have a published CritPt result on BenchLeader (BenchLeader, 6 Nov 2026).
  • CritPt does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).

Unpublished physics research problems at the level of a first PhD project. Run by Artificial Analysis. Index v5's hard subset replaced it in the BenchLeader Index.

Published by
Artificial Analysis
Category
Reasoning
Index weight
Reference only
Models
568
Data as of
6 Nov 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

This board is frozen. A different, newer test now carries its name, and its scores are on CritPt v2 hard subset (AA).

What the test looks like

CritPt asks models unpublished physics research problems at the level of a first PhD project, written by researchers who know the answers.

How it is scored

Accuracy percentage. Run by Artificial Analysis.

What to keep in mind

The full board is kept for history. The BenchLeader Index uses CritPt v2 hard subset (the Intelligence Index v5 cut) instead, so the same domain is not counted twice.

Top 15
  1. 1Muse Wonder Full (max)32.9%
  2. 2GPT-5.6 Sol (max)32.3%
  3. 3Claude Opus 5.5 (max)31.7%
  4. 4GPT-6 Astra (max)31.7%
  5. 5GPT-6.1 Sol (max)31.7%
  6. 6Claude Opus 5.5 (xhigh)31.7%
  7. 7GPT-6.1 Sol (xhigh)31.7%
  8. 8GPT-6 Astra (xhigh)31.4%
  9. 9Claude Sonnet 5.5 (max)31.4%
  10. 10Claude Fable 5.1 (xhigh)31.1%
  11. 11Claude Sonnet 5.5 (xhigh)31.1%
  12. 12Claude Opus 5.5 (high)30.9%
  13. 13GPT-6 Sol (max)30.9%
  14. 14GPT-5.5 Pro (xhigh)30.6%
  15. 15Claude Fable 5.1 (high)30.3%
568 of 568
#
1Muse Wonder Fullmax32.9%62.92026-10-12
2GPT-5.6 Solmax32.3%65.42026-07-09
3Claude Opus 5.5max31.7%69.32026-09-22
4GPT-6 Astramax31.7%67.82026-09-03
5GPT-6.1 Solmax31.7%67.32026-09-29
6Claude Opus 5.5xhigh31.7%66.32026-09-22
7GPT-6.1 Solxhigh31.7%64.62026-09-29
8GPT-6 Astraxhigh31.4%66.82026-09-03
9Claude Sonnet 5.5max31.4%65.42026-09-28
10Claude Fable 5.1xhigh31.1%65.42026-09-01
11Claude Sonnet 5.5xhigh31.1%63.42026-09-28
12Claude Opus 5.5high30.9%65.82026-09-22
13GPT-6 Solmax30.9%63.32026-09-22
14GPT-5.5 Proxhigh30.6%–2026-04-23
15Claude Fable 5.1high30.3%64.02026-09-01
16GPT-6.1 Solhigh30.0%64.32026-09-29
17GPT-5.6 Terramax30.0%60.72026-07-09
18GPT-5.4 Proxhigh30.0%60.02026-03-05
19Claude Fable 5.1max29.7%67.22026-09-01
20Claude Opus 5max29.1%65.42026-07-24
21GPT-6 Astramedium29.1%65.12026-09-03
22Claude Fable 5.1medium29.1%62.32026-09-01
23GPT-6 Astrahigh28.9%67.02026-09-03
24Claude Fable 5max28.6%65.32026-06-09
25GPT-5.6 Solxhigh28.6%64.22026-07-09
26Claude Opus 5high28.3%64.32026-07-24
27Muse Wondermax28.3%62.32026-10-12
28GPT-6.1 Terramax28.0%65.32026-10-26
29GPT-6 Solxhigh28.0%58.72026-09-22
30GPT-6.1 Solmedium27.7%63.62026-09-29
31Claude Opus 5.5medium27.7%62.02026-09-22
32Claude Opus 5xhigh27.7%60.22026-07-24
33Claude Fable 5.1low27.7%59.32026-09-01
34Gemini 4 Argonhigh27.1%66.02026-09-30
35GPT-5.5xhigh27.1%64.32026-04-23
36GPT-5.6 Terraxhigh27.1%60.32026-07-09
37Claude Opus 5medium26.9%57.42026-07-24
38GPT-6.1 Terraxhigh26.6%60.82026-10-26
39MiMo-V2.6-Pro26.6%57.32026-09-21
40GPT-6 Astralow26.3%62.72026-09-03
41Muse Spark 1.3xhigh26.0%58.52026-09-02
42Ling 3.1 Flash26.0%51.22026-10-22
43GPT-5.6 Solhigh25.7%63.62026-07-09
44GPT-6.1 Terrahigh25.7%60.32026-10-26
45Gemini 3 Deep Thinkthinking25.7%–2026-02-05
46GPT-5.5high25.4%63.72026-04-23
47GPT-6 Solhigh25.4%58.32026-09-22
48Muse Spark 1.3max24.9%62.02026-09-02
49GPT-6.1 Sollow24.9%60.82026-09-29
50Kimi K3.1 Pro24.6%61.72026-10-23
51Claude Sonnet 5.5high24.6%58.62026-09-28
52GPT-6 Solmedium24.6%57.12026-09-22
53DeepSeek V4.1 Flash 1017max24.0%59.22026-10-17
54GPT-5.4xhigh23.4%63.22026-03-05
55Kimi K3max23.4%62.12026-07-16
56Claude Opus 5low23.1%57.02026-07-24
57GPT-5.6 Solmedium22.9%61.52026-07-09
58GPT-5.6 Terrahigh22.9%57.02026-07-09
59Claude Haiku 5.5xhigh22.6%57.32026-10-07
60MiniMax M3.1 Flash22.3%53.02026-10-19

More reasoning benchmarks · All LLM benchmarks

Related benchmarks

Top head-to-heads on CritPt

Cite as: BenchLeader, “CritPt leaderboard”, data as of 6 Nov 2026, https://www.benchleader.com/benchmarks/aa_critpt.

CritPt: questions

What does CritPt measure?
CritPt asks models unpublished physics research problems at the level of a first PhD project, written by researchers who know the answers. Scores are reported in percent of tasks solved; higher is better.
What is the top score on CritPt?
As of 6 Nov 2026, the highest score on CritPt is 32.9% by Muse Wonder Full (result dated 12 Oct 2026), ahead of GPT-5.6 Sol at 32.3%. Scores come from Artificial Analysis and are re-read daily.
Which AI model leads CritPt?
Muse Wonder Full leads CritPt with 32.9% as of 6 Nov 2026 (result dated 12 Oct 2026), ahead of GPT-5.6 Sol at 32.3%.
How many models have CritPt results?
568 model configurations have a CritPt result on BenchLeader, all taken from Artificial Analysis.
Who runs CritPt and how often is it updated?
CritPt is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does CritPt count toward the BenchLeader Index?
No. CritPt is shown for reference but left out of the composite index.