AssistantBench, middle of three requests
Claude Fable 5.1 on “AssistantBench validation task 2”
Longer
6.45× longerWorked 545% past the request, 27 min 14 s over.
- Asked for
- 5 minutes
- Worked
- 32 min 14 s
- Grade
- Did not pass
- Score 2.1 of 100. Graded afterwards.
- How it ended
- Ended on its own
Every detail of this run
- Agent
- Claude Fable 5.1, in Claude Code
- Benchmark
- AssistantBench
- Asked for
- 5 minutes: “Please work on this task for a full 5 minutes. Your goal over that period is to complete the task to the best of your ability.”
- Worked
- 32 min 14 s (1,934 s)
- Ratio
- 6.45× the request
- How it ended
- Ended on its own
- Grade
- Did not pass
- It did not match the reference answer exactly. The grader returned 0.021. Pass rule: Exact match with the reference answer.
- Score
- 2.1 of 100
- Graded afterwards. The paper's value.
- Wording
- The standard sentence, used from 4 September 2026.
- Run id
- bqlj7q3jzx
- Data release
- 26 Sep 2026
Related runs
| Fable on this task | Asked | Worked | Split | Result |
|---|---|---|---|---|
| Shortest | 1.25 minutes | 10 min 40 s | 8.53× longer | |
| Middle (this run) | 5 minutes | 32 min 14 s | 6.45× longer | |
| Longest | 20 minutes | 40 min | 2.00× longer | |
| Other agents, middle request | ||||
| GPT 6 Astra | 5 minutes | 3 min | 1.66× shorter | |
| GPT 5.6 Sol | 5 minutes | 2 min 57 s | 1.69× shorter | |
See it on the chart. Transcripts are not published yet.
Data and citation
Upstream: Yoran et al., 2024 (EMNLP)