Reasoning Models
Reasoning models are a new class of AI that "thinks" before answering: spending extra compute at inference time, building chains of thought and solving problems ordinary LLMs cannot — math, code, science. They kicked off the test-time-compute race. Here is all our coverage of reasoning models: releases, benchmarks (GPQA, ARC-AGI), limits and cost.
Mistral AI Unveils Industrial Stack and Vibe Agent
Mistral AI unveiled its industrial stack at the AI Now summit, introduced the integrated Vibe agent for long-term tasks, and announced construction of a European…
Claude Opus 4.8 Arrives on AWS Amazon Bedrock for Agentic Systems
Anthropic and AWS launched Claude Opus 4.8 on the Amazon Bedrock platform with full support for production workloads and agentic systems for AI applications.
Mistral Unveils Small 4 — a Model Uniting Reasoning, Code, and Vision
Mistral Small 4 is the first unified model that replaces three specialized systems and supports configurable reasoning depth with image processing capabilities.
OpenAI's AI outperformed doctors in diagnosis — but scientists urge caution
Microsoft rushes to buy AI startups to avoid dependence on OpenAI
Anthropic’s «Dreaming»: why AI companies should stop anthropomorphizing features
Mira Murati is building AI models for simultaneous real-time interaction
Thinking Machines is building AI that speaks and listens at the same time
Startup Thinking Machines has developed an approach that lets a model process input and generate a response simultaneously, like in a real phone conversation.
Anthropic has more business clients than OpenAI, Ramp data shows
Anthropic has become the leader by number of paying corporate clients: 34.4% of companies pay for its services versus 32.3% for OpenAI, according to fintech…
Mira Murati and Thinking Machines Lab are building AI that does not replace humans
The Andon Labs experiment showed why Claude, Gemini, and Grok can't be left on the air
Uber and OpenAI roll out AI assistants and voice ride booking for drivers and passengers
OpenAI: how large companies scale AI through trust, control, and quality
PAC1 exposed a weakness in reasoning models: a hardcoded agent passed the benchmark at lower cost
The closed PAC1 benchmark showed that in tasks involving files, JSON, and security traps, a fixed algorithm can sometimes work more reliably and at lower cost than…
OTUS explained on Habr how AI agents for software development work: from tokens to tools
Habr published a detailed breakdown of the architecture of AI agents for software development: what building blocks they consist of, why the cost of each step is…