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.
AWS published 6 approaches to optimizing MCP tools to save context
AWS examined MCP tool design problems: tool definitions overflow the context window, and models get confused. It proposed 6 solutions — from renaming parameters to…
Apple identified when on-policy distillation helps model training
Apple ML researchers identified the conditions under which on-policy distillation is effective for training reasoning models, and when it may be ineffective or even…
GRAPHEVAL: New Graph Framework Reveals That Self-Consistency Masks Reasoning Errors in LLMs
Researchers proposed GRAPHEVAL—a graph framework that identifies logical errors in language models where standard Self-Consistency overlooks them, seeing only…
Neural networks are bad calculators: why LLMs fail at arithmetic and what to do about it
SkyPilot and Hugging Face launched storage with no egress fees
Former Qwen Technical Lead: Why Hybrid Reasoning Failed and What's Next
Humanity's Last Exam: Why the Top AI Benchmark from CAIS Is Called a Distraction
Apple Introduced Conformal Thinking — Risk Management for Reasoning Models Without Extra Tokens
Apple ML Research presented the Conformal Thinking framework: a method that automatically manages token budgets for reasoning models, guaranteeing specified error…
LangChain Prompt Caching: Deep Agents Cut LLM Costs Up to 80%
Deep Agents in LangChain automatically caches repeated parts of the prompt and cuts token costs by up to 80% across all major providers — with no additional setup.
Hugging Face compares all LoRA alternatives: who wins at fine-tuning LLMs
Anthropic gave opposite advice for Fable 5 and Opus 4.8: what to change in prompts
The Claude Code team explains what changes when code is no longer scarce
Claude Haiku, Gemini Flash, and GPT-4o mini are changing the economics of AI — flagship models no longer have a monopoly
OpenAI unveils GPT-5.6 with three tiers — Sol, Terra, and Luna — and reasoning modes
OpenAI has announced the GPT-5.6 family with three models — Sol, Terra, and Luna — and new reasoning modes, max and ultra. Access is currently limited to selected…
Trump asked OpenAI to delay the GPT-5.6 release: the model will go only to select partners
The Trump administration demanded that OpenAI not release GPT-5.6 to the public — the new model will go only to a select group of the company's trusted partners.