Skip to content

Search stories, models, companies…Subscribe

Topic · Concept

Reasoning models

Reasoning models are language models trained to work through a problem in intermediate steps, sometimes called "thinking," before giving a final answer. The approach improves results on math, coding and multi-step tasks, but the extra tokens add cost and latency.

11 stories · updated Oct 27

Follow via RSS Get Reasoning models alerts

Key facts

Checked Oct 9

Chain-of-thought prompting: Wei et al., Jan 28, 2022
OpenAI o1 / "Learning to reason with LLMs": September 12, 2024; RL on chain of thought; scales with test-time compute
DeepSeek-R1: Jan 22, 2025; reasoning via reinforcement learning
Claude thinking controls: budget_tokens (extended thinking): Claude 4.5 and earlier; deprecated on 4.6; rejected (400 error) on 4.7 and later, incl. the 5.5 models and Fable 5.1, which use adaptive thinking plus effort. Thinking counts toward max_tokens and billed output tokens

Start here

Models · ExplainerReasoning models explained: thinking tokens, what they cost, and when to use themClaude, GPT-6 and Gemini all bill hidden thinking as output tokens. How each vendor exposes it, what it costs in practice, and when to turn it down.Oct 9·7 min read

Latest on Reasoning models

Policy · NewsAI homework grading at Berkeley: Fields medalist Richard Borcherds stops grading homework for correctness, seeing 'no point' because of AIAI homework grading at Berkeley: Fields medalist Richard Borcherds told the Daily Californian he now grades homework only on completion, seeing 'no point' in paying readers to check correctness once students use AI. What he said, what Berkeley's AI policy says, and what surveys show about student AI use.Oct 27·6 min read

Models · CompareGPT-6 Astra vs GPT-6.1 Sol vs GPT-6 Luna: what each OpenAI model is for, prices, limits and which to use (October 2026)GPT-6 Astra vs 6.1 Sol vs Luna, compared on OpenAI's own pages as of October 23, 2026: Astra costs $10/$50 per million tokens, GPT-6.1 Sol $2/$10 and Luna $0.10/$0.50, all with a 1.05M-token context, but they sit in different parts of ChatGPT.Oct 24·Checked Oct 23

Research · NewsKLS conjecture proof: three papers claim the 31-year-old problem, and what AI did in eachThe KLS conjecture proof arrived in early October 2026 from more than one team. Microsoft's Xinzhi Zhang and Zhao Song credit a human-directed AI agent search; Bizeul, Klartag and Lehec say most of their proofs were found by ChatGPT. None is peer reviewed yet.Oct 24·6 min read

Open source · NewsDeepSeek V4.1 Flash 1017: what's new, parameters, price and VRAM requirementsDeepSeek V4.1 Flash 1017 arrived on October 17, 2026 as a minor update to September's V4.1-Flash: same 763B-parameter MIT-licensed weights format, same API price, better vendor-reported scores, and a new paid Search and Fetch API. Here's what changed, and what the VRAM requirements question really comes down to.Oct 18·6 min read

Research · AnalysisGLM-5.3 update-aware reasoning: what Apollo Research found when Z.ai's model reasoned about its own trainingApollo Research reported on October 15, 2026 that Z.ai's GLM-5.3 sometimes reasons, unprompted, about how its actions could become training data and shape future versions of itself. Here's what the GLM-5.3 update-aware reasoning finding shows, how rare it was, how Apollo measured it, and what it doesn't prove.Oct 18·6 min read

Models · ExplainerAnthropic's distillation defenses explained: preserved thinking, extraction classifiers and account bansOn October 12, 2026, Anthropic said new defenses against distillation have cut the data harvested from Claude by around 90 to 99%. What distillation is, why reasoning transcripts are the prize, how preserved thinking and the new classifiers work, and what they change for developers.Oct 14·7 min read

Models · ExplainerOpenAI loop transformers: what's reported about GPT-6.1 Sol and OpenAI's next model, and what a looped transformer isThe Information reported on October 13, 2026, that OpenAI is working through safety issues tied to loop transformers before shipping its next flagship model. What's actually confirmed, what OpenAI's GPT-6.1 Sol safety addendum shows, how the cancelled GPT-6.1 Astra fits in, and how looped transformers work.Oct 14·7 min read

Developers · GuideHow to use Gemini 4 Argon: in the Gemini app, AI Studio, the Gemini API and AntigravityHow to use Gemini 4 Argon depends on where you want it: AI Pro or Ultra for the Gemini app, Ultra for AI Studio's Playground and Antigravity, and a billing-enabled key for the API. Plans, steps from Google's docs, sample code, costs, limits and fixes.Oct 14·8 min read

Models · NewsGPT-6 reaches every ChatGPT user: what OpenAI shipped, what it costs, and where each model runsFour GPT-6 models in five weeks, split across ChatGPT's Chat and Work tabs, Codex and the API. Here's the verified map, with prices.Oct 9·6 min read

Research · AnalysisOpenAI's Navier–Stokes claim, explained for builders: the verifier is the storyOpenAI says 10,000 AI agents found a blowup in the Navier–Stokes equations and proved it in Lean. What is checked, what isn't, and what it means for agent builders.Oct 9·6 min read

Nothing in this format on this page — try another filter or the next page.