General baseline
BIG-bench
200-plus tasks designed to exceed model capability at release. Useful for finding remaining failure types, not average workflow performance.
As of 13 August 2026. Conditional general baseline. Srivastava et al. A failure-frontier map, not average workflow performance.
BIG-bench is 200-plus tasks. Task metrics stay separate. A single BIG-bench score is refused here as a ranking device.
Data is not a single download. Recheck per-task terms. No Fable 5 or GPT-5.6 table is cited.
- Scale
- 200-plus tasks. Data is not a single parquet download.
- Provenance
- Srivastava et al. Recheck per-task terms.
- Contamination
- High. Public since 2022.
- Cannot establish
- A professional-document score or a single BIG-bench winner.
Metrics, kept separate
- Task-specific official metrics
No cited public figure is installed for this pack. Method, scale, and gates still apply. A missing chart is not a pass rate.
What the figures bound
Use the pack, if at all, to find remaining failure types. Do not treat it as document work.
No chart is installed.
What this page does not conclude
- A professional-document score
- A single BIG-bench winner
- GPT-5.6 or Fable 5 on this suite
- A diligence-product ranking
Cited as calibration on these memos — not a score of those products.
Not attached to a software, advisory, or automation memo yet.