Skip to content

Reasoning

Progress in model reasoning: chain of thought, reasoning models, math and logic benchmarks and the debates around them.

Latest picks

21–40 of 696

Nov 10Tuesday

Simon Willison

llm-anthropic 0.31 adds support for Fable 5.2 and Mythos 5.2

Simon Willison released llm-anthropic 0.31, adding Anthropic's Fable 5.2 and Mythos 5.2, both launched today, so the new models can be called from the CLI.

Why it matters: Willison tested Fable 5.2's five reasoning levels; at max it ran out of tokens without producing an image, while the two earlier versions succeeded at the same level.

TechCrunch · AI

Anthropic releases Mythos 5.2 and Fable 5.2, raising concerns about recursive self-improvement

Anthropic released Mythos 5.2 and Fable 5.2 on Monday, citing gains in coding, reasoning, vision and communication. Fable's input price drops from $10 to $8 per million tokens, responses are 30% faster at every effort level, and Enterprise Frontier Safeguards support zero data retention (ZDR).

Why it matters: Fable's input price falls from $10 to $8 per million tokens, making Fable 5.2 cheaper than main rival GPT-6 Astra.

Hacker News front page

Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, saying both improve on coding, reasoning and vision, with output up to 30% faster than version 5.1.

Why it matters: Anthropic gives one underlying model different safeguards and access terms, so general users and cyber-security or biomedical research institutions get clearly different task permissions.

The Verge · AI

Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2

Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2, citing gains in coding, reasoning and vision and faster output than version 5.1. In pre-release tests, the company says, the two models helped complete software migrations, designed 3 working electronic devices from scratch, solved 2 open problems in black hole physics and formalized a proof of Fermat's Last Theorem within 30 hours.

Why it matters: Anthropic set different access scopes for the two models: Fable is live today, while Mythos is initially limited to vetted organizations.

Computing Life · Share · Yage

Agent harness techniques tested: which framework designs raise scores and which just raise cost

Controlled experiments from Surge AI and FireFlash show coding agents can improve through deeper single-model reasoning, test and goal-check gates, and a reworked terminal pipeline. In Surge AI's run, Compound Engineering lifted cost to 8x baseline while the score moved only from 31.1 to 32.9. Claude Opus 5.5 with higher thinking settings reached 42.4 at 3.5x baseline cost.

Why it matters: The two teams break down the cost and payoff of agent harness techniques, comparing multi-agent collaboration, deeper single-model thinking and terminal pipeline changes.

Nov 6Friday

Latent Space

Thinking Machines releases Inkling 2 Small: 68-hour pretrain, leads web dev evals over Muse Spark Open

Thinking Machines released Inkling 2 Small, a multimodal MoE with 594B total parameters and 17B active. It is available on Tinker for fine-tuning and inference.

Why it matters: Inkling 2 Small's higher AI benchmark scores come with heavier output token use, but its low unit price keeps per-task cost below the comparison model.

Hacker News front page

Tencent releases Hy4 with vision understanding and computer operation, plans to open model weights

Tencent released Hy4, a new large language model that adds vision understanding and computer operation. The company says it will publish the model weights and a system card in the coming days. Hy4 improves instruction following, long-context understanding and long-horizon agent tasks, and is already live in WorkBuddy, CodeBuddy and Tencent Cloud TokenHub.

Why it matters: Hy4 adds vision and computer operation while keeping Hy4 Preview pricing, so the wider task range comes without a higher standard call price.

Nov 3Tuesday

Hacker News front page

Astra edits 0.016% of Qwen3.5-0.8B's weights, lifting MATH-500 from 7.9% to 48.1%

A Weight-Space Programming research report says GPT-6 Astra directly edits 0.016% of Qwen3.5-0.8B's weights, raising MATH-500 from 7.9% to 48.1%. The edit eases reasoning loops in Qwen; on Gemma, which rarely loops, the gain is limited.

Why it matters: The Qwen edit eases reasoning loops while Gemma barely moves, separating removal of a behavior block from added math ability.

Google Blog · Gemini

Google Recaps October 2026 AI Updates, Covering Gemini 4 and Genie 4

Google's October 2026 AI recap covers the global rollout of Gemini 4 Argon, the announcement of Gemini 4 Carbon, and the world model Genie 4. Argon goes out worldwide while Carbon stays in limited testing.

Why it matters: Google is running Argon's global release and Carbon's limited test on separate tracks, so the two frontier models differ in access and safety-testing stage.

Nov 2Monday

Computing Life · Share · Yage

Why LLM inference splits prefill and decode, and how that leaked user code at Cloudflare

Prefill-decode disaggregation speeds up inference by splitting the two stages, but Cloudflare's GLM-5.3 deployment leaked cache across tenants through a request-handling bug. Its postmortem says 155 long-context replies mixed in other users' conversations or code: over-limit requests were replaced with a one-token placeholder yet still crossed nodes, and decode nodes read stale cache that had not been overwritten.

Why it matters: Cloudflare's fix adds transfer-count checks and delayed memory reclamation, folding data overwrite and page ownership into safety checks when nodes exit abnormally.

Nov 1Sunday

Computing Life · Share · Yage

Reproducing the ETH conjecture counterexample with Opus 5.5 for $1.22

Duck Ge reports reproducing a counterexample to the ETH conjecture for $1.22, calling claude-opus-5-5 through Claude Code with a public five-sentence prompt. With all tools and web search off, the run took 8 minutes 41 seconds and 48,101 thinking tokens, and output a five-section proof. The post explains the prompt's algorithm and stresses the result does not mean P=NP.

Why it matters: A hands-on reproduction turns into a prompt-writing method: give the model counterintuitive paths, high-level concepts and worst-case bottlenecks.

Oct 30Friday

The Decoder

Epoch AI: GLM 5.3 builds full Firefox exploit chains in under 48 hours with no human help

Epoch AI found that the open-weight model GLM 5.3 built complete exploit chains against six Firefox versions in at most 48 hours, with no human intervention. It exploited the gap between a patch going public and reaching users.

Why it matters: Epoch AI shows how GLM 5.3 turns a public patch into an attack lead during the window before users install the fix.

Oct 29Thursday

Mistral AI

Mistral releases open weights for Mistral Large 4 (Le Chonk), a 1-trillion-parameter multimodal model

Mistral AI released the open weights of an improved Mistral Large 4 (Le Chonk). The 1-trillion-parameter native multimodal model improves coding, defensive cybersecurity and reasoning efficiency. Self-hosting and customization are allowed, but companies running model-as-a-service need a separate commercial license.

Why it matters: The open weights allow self-hosting and customization, while commercial model-as-a-service use needs a separate license, so the boundary between self-deployment and commercial use is explicit.

Hacker News front page

Google's Gemini 4 Carbon tops Artificial Analysis Intelligence Index v5

Google's Gemini 4 Carbon scored 59 on the Artificial Analysis Intelligence Index v5, taking first place. It ranked second on the coding agent index. Artificial Analysis compared it with similar models and found its cost per task lower than Claude Opus 5.5.

Why it matters: The benchmark puts Gemini 4 Carbon first on general intelligence and cheaper per task than Claude Opus 5.5.

Google Blog · Gemini

Google releases Gemini 4 Carbon, targeting cybersecurity and long-horizon agent tasks

Google announced Gemini 4 Carbon, with gains in cyber defense, coding and long-horizon agent tasks. In external evaluations run without safeguards enabled, it scored 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploiting them and 71.0% on CWE-bench for fixing them.

Why it matters: Google kept the previous generation's entry price for Gemini 4 Carbon and reports secure code generation succeeding more than 2x as often as before.

Oct 28Wednesday

TechCrunch · AI

OpenAI and Anthropic claim to refute the Exponential Time Hypothesis, and computer science is asked to check

OpenAI and Anthropic claim their models found algorithms that solve certain logic problems in sub-exponential time, which could refute the Exponential Time Hypothesis (ETH). ETH is related to P vs NP, but refuting it does not mean P = NP has been proven. MIT professor Ryan Williams is organizing academic review of the algorithms and pushing discussion of research that rests on the assumption.

Why it matters: If the claim holds, computer scientists would have to re-examine research built on the Exponential Time Hypothesis.

Oct 27Tuesday

The Decoder

OpenAI taps outside experts, including a Nobel laureate, to review AI-generated math and physics proofs

OpenAI has invited outside experts, among them Nobel physics laureate David Gross, to review AI-generated proofs in math, computer science and physics. The company says the work includes disproving the exponential time hypothesis, solving a local smoothing conjecture in all dimensions, and establishing the mathematical existence of several four-dimensional quantum field theories. The Yang-Mills result covers existence only, not the mass gap, so it is not a full solution of that Millennium Prize problem.

Why it matters: OpenAI is routing AI research results in three fields through separate review paths, with formal checks and expert review covering logic and meaning respectively.