BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases8

CritPt v2 hard subset (AA)

As of 28 Oct 2026, the top score on CritPt v2 hard subset (AA) is 71.5% by GPT-6 Astra (OpenAI), ahead of GPT-6 Astra at 71.0%, across 86 model configurations with a published result (Artificial Analysis, result dated 3 Sept 2026). CritPt v2 hard subset: 40 private physics research problems from CritPt v2, used in Artificial Analysis' Intelligence Index v5, with a code-execution tool. CritPt v2 hard subset (AA) counts toward the BenchLeader reasoning score.

Key facts (as of 28 Oct 2026)

  • GPT-6 Astra holds the top CritPt v2 hard subset (AA) score at 71.5% (Artificial Analysis, 3 Sept 2026).
  • GPT-6 Astra is second on CritPt v2 hard subset (AA) at 71.0% (Artificial Analysis, 3 Sept 2026).
  • CritPt v2 hard subset (AA): Private 40-question hard subset of CritPt v2 used in Intelligence Index v5, with a code-execution tool. Run by Artificial Analysis (Artificial Analysis).
  • 86 model configurations have a published CritPt v2 hard subset (AA) result on BenchLeader (BenchLeader, 28 Oct 2026).
  • CritPt v2 hard subset (AA) feeds the reasoning category of the BenchLeader Index (BenchLeader methodology).

Private 40-question hard subset of CritPt v2 used in Intelligence Index v5, with a code-execution tool. Run by Artificial Analysis.

Published by
Artificial Analysis
Category
Reasoning
Index weight
1.0
Models
86
Data as of
28 Oct 2026
Newest result
26 Oct 2026

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

Not comparable with the earlier test of the same name, whose final scores stay on CritPt.

What the test looks like

CritPt v2 hard subset: 40 private physics research problems from CritPt v2, used in Artificial Analysis' Intelligence Index v5, with a code-execution tool.

How it is scored

Accuracy percentage. Run by Artificial Analysis.

What to keep in mind

Private hard subset — not the same as the full CritPt board Artificial Analysis also publishes. This is the CritPt result counted in the BenchLeader Index.

Top 15
  1. 1GPT-6 Astra (high)71.5%
  2. 2GPT-6 Astra (medium)71.0%
  3. 3GPT-6 Astra (max)70.5%
  4. 4GPT-6 Astra (xhigh)70.5%
  5. 5GPT-6.1 Sol (max)68.5%
  6. 6GPT-6.1 Sol (xhigh)68.0%
  7. 7GPT-6.1 Sol (high)67.5%
  8. 8GPT-6.1 Sol (medium)65.0%
  9. 9GPT-6.1 Terra (max)64.5%
  10. 10Claude Opus 5.5 (xhigh)63.0%
  11. 11Claude Opus 5.5 (max)61.0%
  12. 12Claude Fable 5.1 (high)60.5%
  13. 13GPT-6.1 Terra (high)60.5%
  14. 14GPT-6 Astra (low)60.0%
  15. 15GPT-6.1 Sol (low)59.5%
86 of 86
#
1GPT-6 Astrahigh71.5%67.32026-09-03
2GPT-6 Astramedium71.0%65.22026-09-03
3GPT-6 Astramax70.5%68.32026-09-03
4GPT-6 Astraxhigh70.5%67.12026-09-03
5GPT-6.1 Solmax68.5%67.62026-09-29
6GPT-6.1 Solxhigh68.0%64.82026-09-29
7GPT-6.1 Solhigh67.5%64.32026-09-29
8GPT-6.1 Solmedium65.0%63.12026-09-29
9NewGPT-6.1 Terramax64.5%65.02026-10-26
10Claude Opus 5.5xhigh63.0%66.82026-09-22
11Claude Opus 5.5max61.0%69.82026-09-22
12Claude Fable 5.1high60.5%63.52026-09-01
13NewGPT-6.1 Terrahigh60.5%59.92026-10-26
14GPT-6 Astralow60.0%62.02026-09-03
15GPT-6.1 Sollow59.5%59.02026-09-29
16Claude Opus 5.5high58.0%65.92026-09-22
17NewGPT-6.1 Terraxhigh57.5%60.42026-10-26
18Claude Fable 5.1xhigh56.5%65.02026-09-01
19Claude Fable 5.1medium56.5%61.52026-09-01
20Claude Sonnet 5.5max56.0%65.72026-09-28
21GPT-5.6 Solmax56.0%65.42026-07-09
22NewGPT-6.1 Terramedium56.0%58.32026-10-26
23Claude Sonnet 5.5xhigh54.0%63.42026-09-28
24Muse Wonder Fullmax53.0%62.72026-10-12
25Gemini 4 Argonhigh52.5%66.02026-09-30
26Claude Opus 5.5medium52.5%62.22026-09-22
27Claude Fable 5.1max51.0%66.92026-09-01
28NewGPT-6.1 Terralow48.5%54.92026-10-26
29Muse Wondermax48.0%61.22026-10-12
30Claude Fable 5.1low47.5%58.12026-09-01
31Claude Sonnet 5.5high47.5%57.92026-09-28
32Muse Spark 1.3max46.0%60.92026-09-02
33NewKimi K3.1 Pro45.0%59.82026-10-23
34Claude Haiku 5.5xhigh44.5%56.02026-10-07
35Gemini 3.8 Flashmedium44.0%58.62026-09-02
36Gemini 3.8 Flashhigh43.5%60.32026-09-02
37Claude Haiku 5.5high42.5%54.52026-10-07
38GPT-6 Lunamax41.5%55.42026-09-22
39Claude Haiku 5.5max41.0%56.42026-10-07
40Grok 4.7low41.0%54.42026-09-21
41Kimi K3max39.5%62.22026-07-16
42Ling 3.1 Flash37.5%49.82026-10-22
43GPT-6 Lunaxhigh35.5%50.22026-09-22
44Grok 4.7high35.0%56.12026-09-21
45MiMo-V2.6-Pro34.0%56.92026-09-21
46Claude Opus 5.5low33.5%57.82026-09-22
47GLM 5.3max33.5%57.62026-08-18
48Step 5 Preview32.5%55.02026-09-18
49NewGPT-6.1 Terrano reasoning32.5%51.62026-10-26
50Grok 4.7xhigh32.0%57.42026-09-21
51Gemini 3.8 Flashlow31.5%53.72026-09-02
52MiniMax M3.1 Flash31.0%52.62026-10-19
53GPT-6 Lunahigh30.5%49.22026-09-22
54Claude Haiku 5.5medium30.0%48.52026-10-07
55MiMo-V2.6-Flash29.5%53.82026-09-21
56Qwen3.8 Maxmax28.5%59.12026-09-02
57Claude Sonnet 5.5medium28.5%50.12026-09-28
58Muse Spark Openxhigh27.5%51.72026-10-15
59MiniMax M326.5%54.32026-06-01
60Beam 1.026.5%44.62026-10-16

More reasoning benchmarks · All LLM benchmarks

Cite as: BenchLeader, “CritPt v2 hard subset (AA) leaderboard”, data as of 28 Oct 2026, https://www.benchleader.com/benchmarks/aa_critpt_hard.

CritPt v2 hard subset (AA): questions

What does CritPt v2 hard subset (AA) measure?
CritPt v2 hard subset: 40 private physics research problems from CritPt v2, used in Artificial Analysis' Intelligence Index v5, with a code-execution tool. Scores are reported in percent of tasks solved; higher is better.
What is the top score on CritPt v2 hard subset (AA)?
As of 28 Oct 2026, the highest score on CritPt v2 hard subset (AA) is 71.5% by GPT-6 Astra (result dated 3 Sept 2026), ahead of GPT-6 Astra at 71.0%. Scores come from Artificial Analysis and are re-read daily.
Which AI model leads CritPt v2 hard subset (AA)?
GPT-6 Astra leads CritPt v2 hard subset (AA) with 71.5% as of 28 Oct 2026, ahead of GPT-6 Astra at 71.0% (result dated 3 Sept 2026).
How many models have CritPt v2 hard subset (AA) results?
86 model configurations have a CritPt v2 hard subset (AA) result on BenchLeader, all taken from Artificial Analysis.
Who runs CritPt v2 hard subset (AA) and how often is it updated?
CritPt v2 hard subset (AA) is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does CritPt v2 hard subset (AA) count toward the BenchLeader Index?
Yes. CritPt v2 hard subset (AA) contributes to the reasoning category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.