Daily AI CatchupAI Catchup — Wednesday, October 7, 2026
Wednesday, October 7, 2026Reasoning
OpenAI Releases 722 Math Papers From Frontier Model, Advances Mathematical Reasoning
OpenAI published a trove of mathematical proofs from an internal model, complete with Lean formalizations and reasoning traces.
LlmsAnthropic's Claude Haiku 5.5 Cuts Costs 75%, Adds Adjustable Reasoning
Claude Haiku 5.5 slashes pricing by three-quarters while adding configurable reasoning modes for a cheaper, faster reasoning option.
ApiOpenAI Launches Decisions API: 10x Faster Routing for Production Decision-Making
OpenAI's Decisions API makes routing, classification, and scoring decisions 10x faster than language model alternatives at just $0.10 per million tokens.
Frontier-ModelsMistral Large 4: One-Trillion-Parameter European AI Model Arrives
Mistral shipped a trillion-parameter multimodal model positioned to compete with OpenAI and closed-source leaders while keeping the EU in the AI race.
BenchmarksDeepSeek V4.1 Flash Surpasses US Models on Agentic Coding Benchmarks
A Chinese AI model now leads on agentic coding tasks, narrowing the US-China frontier gap to just 3%, intensifying competition.