Skip to content

Benchmark

Agents' Last Exam · Overall Score

Agentic benchmarks measure how well models carry out tasks that take several steps. Higher scores are better.

Models and systems measured
17
Scores recorded
50
Sources
1
IQ weight
0.79

Score distribution

49
Mean
51.2
Median
7.4
Std dev
38.5
10th percentile
56.4
90th percentile
29.6
Min
29.659.3
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.79 for chat models, measured over 6 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model or systemScore
1GPT-6 AstraOpenAI launch table · highest published effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.59.3
2GPT-6 Astramax effort · $6.2337 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.59.3
3GPT-6 Astraxhigh effort · $5.397 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.58.3
4GPT-6 Astrahigh effort · $4.6396 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.57.8
5GPT-6 Astramedium effort · $4.1046 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.57.6
6GPT-6 Solmax effort · $2.9313 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.56.4
7Claude Opus 5high effort · $7.2917 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.9
8Claude Opus 5xhigh effort · $10.0226 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.5
9Claude Opus 5OpenAI launch table · highest published effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.5
10GPT-6 Solxhigh effort · $1.668 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.4
11GPT-5.6 Solxhigh effort · $5.0764 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.53.6
12GPT-5.6 SolOpenAI launch table · highest published effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.53.6
13GPT-5.6 SolQwen表:Overall Score· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.53.6
14GPT-6 Astralow effort · $3.0726 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.53.4
15GPT-6 Solmedium effort · $1.2664 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.53.1
16Claude Opus 5medium effort · $5.2907 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.53
17GPT-5.6 Solmax effort · $7.1322 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.8
18GPT-5.6 SolOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.7
19Claude Opus 5max effort · $9.7583 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.7
20GPT-6 Solhigh effort · $1.5302 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.6
21Qwen3.8 MaxQwen表:Overall Score· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.4
22GPT-5.6 Solhigh effort · $3.7874 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.4
23GPT-5.6 Solmedium effort · $3.4126 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.1
24Claude Opus 5low effort · $4.0619 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.51.9
25Qwen 3.8 Flash NextQwen3.8-Flash-Next release/model card · Overall Score· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.51.2
26GPT-6 Lunamax effort · $0.1545 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.50.9
27GPT-5.6 TerraOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.50.4
28GPT-5.6 Lunamax effort · $2.5698 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.50.4
29GPT-5.6 LunaOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.50.3
30GPT-5.6 Lunaxhigh effort · $1.5468 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.49.4
31Claude Fable 5OpenAI launch table · highest published effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.48.7
32GPT-6 Sollow effort · $0.8641 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.48.7
33GPT-6 Lunaxhigh effort · $0.1105 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.47.9
34GPT 5.5OpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.46.9
35GPT-6 Lunamedium effort · $0.1054 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.46.8
36GPT-5.6 Lunahigh effort · $0.8389 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.46.1
37Claude Opus 4.8OpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.45.2
38Claude Opus 4.8Qwen表:Overall Score· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.45.1
39GPT-5.6 Sollow effort · $1.6324 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.45.1
40GPT-6 Lunahigh effort · $0.1117 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.43.6
41Qwen 3.8 27BQwen3.8-27B model card · Overall Score· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.42.9
42Claude Fable 5OpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.40.5
43GPT-5.6 Lunamedium effort · $0.375 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.38.5
44GPT-6 Lunalow effort · $0.0247 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.36.3
45Gemini 3.1 Pro PreviewOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.32.1
46Qwen3.7 MaxQwen表:Overall Score· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.31.1
47GPT-5.6 Lunalow effort · $0.1761 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.31
48MiniMax M3.1 FlashMiniMax M3.1 Flash official-weights model-card run · reasoning effort, sampling, and run count not stated for this benchmark · official ALE Overall full-set leaderboard result · CLI-only runs excluded· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.29.6

Not ranked

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
Claude Fable 5 with Opus 4.8 fallbackadaptive effort · $15.2285 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.41.3
Claude Fable 5 with Opus 4.8 fallbackxhigh effort · $28.5542 average cost/task· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.48.7

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.