Bake AI lands two NeurIPS 2026 papers: NARRA-Gym and JobBench with 130 tasks across 35 occupations
my_cat_can_code · x · 2026-09-26
Bake AI announced two environment/benchmark papers accepted at NeurIPS 2026: NARRA-Gym, an executable environment testing whether LLM agents can sustain a coherent, evolving story over many turns while adapting to the user; and JobBench, which evaluates agents on 130 tasks across 35 occupations, targeting the work experts actually want to delegate rather than the most economically valuable tasks. The team says JobBench is already widely used by frontier labs.
Related event: Bake AI's Two Agent Benchmark Papers Accepted by NeurIPS 2026(2 posts)→
More from Research
- Xiaomi open-sources MiMo RL environments repo, individual tasks sell for $100-1000 each — HarveenChadha · 2026-09-26
- Contextual embeddings go mainstream: only Perplexity and Voyage ship them so far — antoine_chaffin · 2026-09-26
- A Curated Paper List for Learning Distributed LLM Training and Inference — East-Muffin-6472 · 2026-09-26
- RL isn't just sharpening existing skills — pre-LLM RL trained from random init — cephaloform · 2026-09-26
- Steelmanning OPSD: train only on tokens after verbatim rule reminders — willcb · 2026-09-26
- Researchers warn Astra reviews obsess over details, miss the big picture — aran_nayebi · 2026-09-26