Loading activity…
@valsai
Public LLM Evaluation // https://vals.ai @8vc @BloombergBeta @pearvc @a16z @HRTVentures @nextladder
AI
ValsAI said Beam 1.0 was the fastest model on the Vals Index and free to the public.
AI
ValsAI says it evaluated 15 models on 1,960 claims paired with 10-K excerpts, with Decider V1.2 leading the decision models.
AI
A user calls the result impressive but says the system is multiple times less efficient than Sol.
AI
ValsAI also puts the model’s cost at $0.64 per task.
AI
ValsAI says the models controlled a simulated robot arm and humanoid using only camera feeds and sensor readings.
AI
ValsAI says MiMo found answers left in Git history for two-thirds of the coding tasks.
AI
A user links to Mercury Decide’s free OpenRouter listing and calls it an “unlimited decision model.”
AI
Mistral says Mistral Large 4 is live via API in public preview now, with open weights due at the end of October. Reuters says the launch is part of Mistral’s push to narrow the gap with leading Chinese open-weight models.
AI
ValsAI says 90-plus Opus 5.5 agents helped uncover YBaMnFeO₅ and KV[Cr(CN)₆] in three days.
AI
Google announces a 1 million-token output limit, up from 64,000, for its new model. Initial access goes to trusted cyber defenders through the Fairwind Program.
AI
ValsAI says a new episode of The Bench discusses owning versus renting AI, capturing institutional knowledge, benchmark limitations and multi-agent systems.
AI
ValsAI says ten agents used Lean to produce a 17,895-line proof in 15 hours. It says the Lean kernel accepted the proof, which identifies a pentagonal bipyramid as the lowest-energy arrangement.
AI
Anthropic says Sonnet 5.5 runs more than 30% faster than Sonnet 5. In pre-release max-effort tests, Artificial Analysis measured per-task costs roughly 50% higher than Sonnet 5’s.
AI
Anthropic says Opus 5.5 performs at Claude Fable 5.1's level for most tasks. Every reports strong visual results from a week of testing, but recommends setting a budget and a clear stopping point.
AI
ValsAI says the run took more than 165 hours of in-game time and two days of real time.
AI
OpenAI says its new Legal Search Index searches U.S. case law, statutes, regulations, court rules and administrative decisions across more than 230 million URLs.
Technology
Tom's Hardware reports that OpenAI's model got further than any other AI system in a 141-hour Minecraft test.
AI
ValsAI claims Astra had gone further than any AI ever had in Minecraft, collecting six blaze rods and three pearls before a creeper blew up its chest of valuables and bed.
AI
Vals backs a market-based approach to independent evaluation and says its benchmarks appear in model cards from every major foundation model lab.
Technology
DeepSeek says its KV cache, which lets the model reuse earlier calculations, needs a quarter of the high-bandwidth memory and an eighth of the SSD storage required by the previous generation's cache.
