Laguna gets 36.8% faster in 24 hours as its benchmark harness is hardened
gajesh · x · 2026-07-28
The team says it made Laguna 36.8% faster internally within 24 hours, and believes there is still substantial performance left to unlock.
It also says the benchmark has been hardened against score gaming by improving the verifier harness. If users find weaknesses or can improve the harness, they are encouraged to reach out, and the benchmark itself may continue to evolve in public.
More from Research
- Goodfire's Nick Cammarata argues InceptionV1 polysemanticity wasn't caused by superposition — petrusenko_max · 2026-10-16
- LLM ideation experiment: just ask for non-obvious ideas — and Opus beats smaller models by a wide margin — arjunrajlab · 2026-10-16
- SimpleMark: multi-bit LLM watermark bounds per-token distortion with f-divergence budget — rishabh16_ · 2026-10-16
- Stanford paper: generative representation learning cuts tokenizer training compute by 7-8x — HirokatuKataoka · 2026-10-16
- Cohere Labs to present confidence estimation for reasoning models at EMNLP 2026 — Cohere_Labs · 2026-10-16
- Vals RSI Index extrapolates Google models hitting 130% autonomous AI R&D by Oct 2027 — HaydnBelfield · 2026-10-16