Laguna says it sped up 36.8% in 24 hours and is opening benchmark improvement
gajesh · x · 2026-07-28
The team says Laguna became 36.8% faster internally within 24 hours, and that there is still more performance to unlock. They are also hardening the benchmark against score gaming.
Beyond that, they are experimenting with continuous open improvement: participants can contribute their agents’ progress, with personal information removed before publication. The stated goal is to let the community discover better research patterns and improve this and future benchmarks in the open.
More from Infra
- A 27B model reaches 24 TPS with on-the-fly 3-bit dequantization on an A6000 — cephaloform · 2026-07-29
- OpenAI's Infra Efficiency Edge Could Enable 10T Parameter Models — haider1 · 2026-07-29
- Red Hat releases an FP8-quantized Kimi-K3 checkpoint tuned for Hopper GPUs — _akhaliq · 2026-07-29
- iFixAi claims to audit deployed AI agents in 120 seconds with 45 checks — socialwithaayan · 2026-07-29
- NVIDIA shows how to run Gemma and Qwen locally on Jetson with Ollama and vLLM — NVIDIA Developer · 2026-07-29
- A joke post imagines phone control of Google’s data centers over brunch — MickeySteamboat · 2026-07-29