Taste-Bench
TasteBench: Multimodal Benchmark for Sensory Prediction, from Molecules to Sustainable Foods
Taste-Bench A taste question shows a frozen agent trajectory stopped at a real decision fork and asks which of two candidate next steps is the better technical decision.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info taste-bench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
2 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Gemini 3.1 Pro | 65.4 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Qwen 3.5 397B | 63.0 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
Taste-Bench
A taste question shows a frozen agent trajectory stopped at a real decision fork and asks which of two candidate next steps is the better technical decision. The candidates are mined from recorded trajectories where attempts diverge at the same point, and the label comes from the later work and its outcome rather than from an expert annotator. The questions cover two domains: software engineering, where agents solve SWE-bench and SWE-bench Pro tasks, and ML research…