9Fable 5Claude Code · Max37.27 ±1.6Fable 5Weighted score37.27 ± 1.65AIME 202511.4Arena Hard54.1GPQA Main28.5*GSM8K83.8HealthBench40.1HumanEval50.8Runs2Time9:05h* Includes fallback scores from Opus 4.8 Max
Base models (zero-shot)8.01Base models (zero-shot)Weighted score8.01AIME 20251.7Arena Hard1.3GPQA Main8.5GSM8K20.4HealthBench9.5HumanEval12.8
PostTrainBench
Agent
Weighted score
AIME 2025
Arena Hard
GPQA Main
GSM8K
HealthBench
HumanEval
Official instruct models, Reference
48.41
29.2
70.2
36.2
87.0
43.3
71.5
Fable 5.1, Claude Code, Max
44.65 ± 2.72
13.2
87.1
29.2 (Includes fallback scores from Opus 5)
86.4
49.6
63.0
Fable 5.2, Claude Code, Max
44.64 ± 1.16
13.7 (Includes fallback scores from Opus 5.5)
73.7 (Includes fallback scores from Opus 5.5)
33.8 (Includes fallback scores from Opus 5.5)
85.2
51.6 (Includes fallback scores from Opus 5.5)
61.3
Opus 5.5, Claude Code, Max
43.80 ± 3.00
11.0
88.1
29.5
72.3
53.0
65.3
GPT-6 Astra, Codex CLI, Max
41.88 ± 0.76
11.8
63.6
33.5
85.9
40.3
69.2
Locus, Intology · Opus 5
41.66 ± 0.82
9.4
66.3
33.1
82.9
44.3
66.6
Gemini 4 Carbon, OpenCode, High
40.98 ± 1.67
12.6
65.5
31.4
85.6
45.3
54.0
Gemini 4 Argon, OpenCode, High
40.44 ± 0.71
9.2
65.5
30.8
83.3
40.9
67.5
GPT-6.1 Sol, Codex CLI, Max
38.79 ± 0.36
10.3
52.8
32.7
86.8
39.9
56.3
Fable 5, Claude Code, Max
37.27 ± 1.65
11.4
54.1
28.5 (Includes fallback scores from Opus 4.8 Max)
83.8
40.1
50.8
GLM 5.3 Flash, Claude Code, Max
36.79 ± 0.11
10.9
63.3
27.3
83.3
37.6
47.1
Opus 5, Claude Code
35.66 ± 2.62
9.7
53.4
31.1
80.5
38.7
40.7
GLM 5.3, Claude Code, Max
34.52 ± 3.39
10.2
38.9
28.3
83.2
38.8
45.6
Kimi K3, Claude Code
31.48 ± 0.63
10.6
34.4
25.5
76.3
29.5
50.1
GPT-5.6 Sol, Codex CLI, Max
31.01 ± 0.20
6.3
26.4
29.6
69.0
27.4
63.3
Opus 4.8, Claude Code, Max
31.00 ± 3.00
7.4
36.1
28.5
69.2
31.7
47.5
Opus 4.8, Claude Code, High
30.66 ± 3.04
8.4
45.8
20.9
69.6
32.4
48.6
GLM 5.2, Claude Code, Max
30.40 ± 1.46
7.4
27.0
30.2
72.6
26.5
52.2
Opus 4.7, Claude Code, xHigh
24.34 ± 1.63
5.9
19.6
27.2
61.0
17.8
40.7
Muse Wonder, OpenCode, xHigh
24.19 ± 1.14
12.3
24.3
26.1
70.6
13.0
24.0
GPT-5.5, Codex CLI, xHigh
23.87 ± 0.48
3.8
12.0
30.6
50.7
19.2
47.0
Grok 4.5, Cursor CLI, High
22.62 ± 0.94
9.6
11.0
26.8
65.0
10.6
35.1
Gemini 3.1 Pro, OpenCode
18.11 ± 1.68
4.0
7.4
17.8
45.3
16.2
37.3
GPT-5.4, Codex CLI, High
17.76 ± 1.80
0.4
9.6
26.3
47.5
12.3
26.7
Base models (zero-shot), Reference
8.01
1.7
1.3
8.5
20.4
9.5
12.8
Official scores come from the PostTrainBench leaderboard and refresh hourly. Entries marked Reported come from the model maker’s own announcement, may use a different setup, and are replaced once the official leaderboard publishes its own run. Scores are not comparable across versions.
