Skip to content

Reasoning

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

Latest picks

81–100 of 696

Oct 8Thursday

Simon Willison

Anthropic ships Claude Haiku 5.5, priced level with GPT-6 Luna on short context

Anthropic released Claude Haiku 5.5, a fast, low-cost model. Up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output tokens; past that threshold the rates rise to $0.50 and $2.50.

Why it matters: Haiku 5.5 matches GPT-6 Luna's price on short context, but the jump past 100,000 tokens and tokenizer differences change the cost comparison.

Hacker News front page

Scott Aaronson on how AI math breakthroughs strain human research and publishing

In a post titled "The Mathocalypse," Scott Aaronson lays out the challenge of understanding, explaining and holding together a research community when AI math results arrive in bulk. He says OpenAI published 372 results yesterday, including a solution to the unique games conjecture, some with Lean certificates, and that humans have understood almost none of the solutions.

Why it matters: Aaronson compares how two companies release AI math results, and how publishing raw solutions versus having scholars write them up splits the explanation work and the academic credit.

Hacker News front page

Scott Aaronson on AI's 'math apocalypse' and the mathematician's new role

Scott Aaronson uses the term 'math apocalypse' for the shock of AI cracking hard math problems, and contrasts OpenAI publishing results directly with Anthropic's path of having mathematicians organize and present them. He says OpenAI released 372 results yesterday, including a solution to the Unique Games Conjecture, some with Lean certificates, yet humans do not understand almost any of the solutions. The piece asks how researchers absorb this work and how the math community keeps exploration and understanding valuable once AI can supply answers.

Why it matters: Aaronson compares how the two companies release AI math results: open drafts set researchers racing to interpret them, while directed collaboration lets the company pick the human explainers.

The Decoder

ChatGPT gets GPT-6, answering with charts, buttons and mini apps

OpenAI is rolling out GPT-6 to ChatGPT users, using an Intelligent UI that turns answers from mostly text into interactive screens with graphics, buttons, charts and forms. The interface adapts to the question, and can build a savings calculator or a game inside the chat. OpenAI says GPT-6 answering while it thinks cuts wait time by 44%, and it scores above GPT-5.6 on an internal hard web search test.

Why it matters: OpenAI is giving ChatGPT an interface that reshapes itself around each question, so comparisons, chart exploration and small tools happen in the conversation.

OpenAI News

OpenAI brings GPT-6 and Intelligent UI to more ChatGPT users

OpenAI released GPT-6 to more ChatGPT users, adding Intelligent UI that combines text, graphics, buttons, forms and interactive tools inside answers. The model can answer first and keep thinking to add more, and in internal evals GPT-6 Instant started answering an average 44% earlier on questions needing web search, compared with GPT-5.6 Instant.

Why it matters: OpenAI folded interface generation into GPT-6's answer flow, so users adjust inputs, explore diagrams and use throwaway tools right in the chat.

Hacker News front page

Anthropic releases Claude Haiku 5.5, cutting average running cost about 75% versus Haiku 4.5

Anthropic released Claude Haiku 5.5 for high-frequency, cost-sensitive work. The company says average running cost is about 75% lower than Haiku 4.5, and the model is available now on all platforms.

Why it matters: Anthropic positions Haiku 5.5 for summarization, compression and subagent tasks, while complex agentic coding still fits Sonnet 5.5 and Opus 5.5 better.

Oct 7Wednesday

Latent Space

OpenAI publishes 722 math papers, says it solved 90 of the first 500 open problems

OpenAI released 722 math papers generated by an unreleased internal model. The accompanying material says the model solved 90 of the first 500 open math problems, including results tied to the Quasi-Riemann Hypothesis.

Why it matters: OpenAI put the papers, argument material and some reasoning summaries in public view, giving outside mathematicians a basis to check each result.

Hacker News front page

OpenAI math repo collects 722 model-generated manuscripts with verification material

OpenAI's math research repository holds 722 manuscripts produced by an unreleased internal model, grouped into 372 research series by related result. The model saw about 4,000 problems during evaluation, and most results came from one shared pipeline, averaging 3 hours of ChatGPT Pro thinking compute per result. Many manuscripts have Lean formalizations, but verification progress varies, and results not yet formalized may have problems. The repo will keep adding formalization material and keeps a revision history.

Why it matters: OpenAI published model-generated math manuscripts with verification material and stated how far each result has been checked, giving context on how reliable these outputs are.

Oct 6Tuesday

The Decoder

Reflection's Beam claims top open-weight performance outside China

Reflection announced Beam, a text model for coding, logical reasoning and AI agent tasks, built around performance at lower compute. The company says it matches GLM 5.2 on hard reasoning tasks while using three to four times less compute, and lets users adjust how much it thinks to trade compute cost against output quality. Beam is still in final safety testing and only some users have an early version. The technical report, developer docs and Apache 2.0 model weights are due later this month.

Why it matters: Reflection claims Beam matches GLM 5.2 on hard reasoning tasks while using a third to a quarter of the compute.

Hacker News front page

Anthropic vs OpenAI subscription review: daily-driver API-equivalent value runs about 5x higher

SemiAnalysis measured subscription quota consumption by token type and found that, on agent workloads, Anthropic's Opus 5.5 plan delivers roughly 5x the API-equivalent value of OpenAI's GPT 6.1 Sol. The comparison breaks plan value down by subscription tier, model and workload combination.

Why it matters: SemiAnalysis separates quota consumption by token type, pushing subscription value comparisons down to specific plan, model and workload combinations.

Oct 3Saturday

AI HOT · Models

Anthropic releases Claude Opus 5.5, first with an ECI score of 167

Anthropic released the closed-weight model Claude Opus 5.5 on September 22, 2026, with an ECI score of 167 that ranks first among 253 tracked models. It scores 77% on the software engineering benchmark MirrorCode and 83% on the multimodal Furniture Assembly Benchmark. Pricing is $4 per 1M input tokens and $20 per 1M output tokens.

Why it matters: Claude Opus 5.5 beats GPT-6 Astra on software engineering but trails it on the hardest math tests.

Hacker News front page

Aleph Alpha releases Kolibri, an open-weight English-German model with 1M-token context

Aleph Alpha released Kolibri, an English-German MoE model with 78B total parameters and about 3B active, supporting up to 1M tokens of context. Full weights are open on Hugging Face under Apache 2.0. Deployment needs the aleph-alpha-inference package with the Kolibri vLLM plugin.

Why it matters: Aleph Alpha turns industry failure cases into a continuously running evaluation and improves the model with synthetic training environments, without training on customer data.

Oct 2Friday

AI HOT · Models

NVIDIA details how Blackwell GPUs accelerate OpenAI's GPT-6 Astra Ultrafast

GPT-6 Astra Ultrafast runs on NVIDIA Blackwell GPUs and generates tokens up to 8x faster than Astra Standard mode. OpenAI used its own models to tune NVIDIA's GPU inference software, and faster responses cut the wait for coding agents that edit, test and debug in loops.

Why it matters: OpenAI tuned NVIDIA's GPU inference software with its own models, and the speedup targets the wait time coding agents hit when they edit, test and debug repeatedly.

Oct 1Thursday

Latent Space

Google DeepMind releases Gemini 4 Argon with 1M output tokens

Google DeepMind released Gemini 4 Argon, aimed at coding, enterprise knowledge work and cyber defense. Access is limited for now to government users and trusted cyber defenders through the Fairwind Program. The company has not said when developers and enterprises will get access.

Why it matters: Gemini 4 Argon took first place on 13 of 19 benchmarks, but access is limited to government users and vetted cyber defenders, with no date for developers and enterprises.

The Decoder

Google releases Gemini 4 Argon: benchmarks match OpenAI and Anthropic, no clear lead

Google released Gemini 4 Argon, its new frontier model. On several benchmarks it matches OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1, but it does not take a clear lead.

Why it matters: Third-party benchmarks put Gemini 4 Argon level with GPT-6 Astra and Claude Fable 5.1, though it burns about twice the output tokens per task.

AI HOT · Models

Artificial Analysis rates Gemini 4 Argon: Google back in the top three on intelligence

Artificial Analysis' evaluation shows Google DeepMind's Gemini 4 Argon scoring 53 on the Artificial Analysis Intelligence Index in high-reasoning mode, level with GPT-6 Astra (max, 53) and one point ahead of GPT-6.1 Sol (max, 52).

Why it matters: Artificial Analysis scored Gemini 4 Argon's high-reasoning mode at 53 on its intelligence index, tied with GPT-6 Astra, with the lowest hallucination rate among comparable models.

Google Blog · Gemini

Google releases Gemini 4 Argon, first to trusted cyber defense teams

Google announced a new model, Gemini 4 Argon, rolling out gradually through the Fairwind Program to selected trusted cyber defense teams. It targets coding, enterprise knowledge work and cyber defense.

Why it matters: Google raised Gemini 4 Argon's output limit from 64K to 1M tokens, giving complex tasks more room for long single-pass reasoning.

Sep 30Wednesday

AI HOT · Models

GPT-6.1 Sol replaces GPT-6 Sol 7 days after launch, 1 point behind GPT-6 Astra on intelligence index

GPT-6.1 Sol replaced GPT-6 Sol 7 days after launch. Its intelligence index is 1 point below GPT-6 Astra, and pricing stays at $2 per million input tokens and $10 per million output tokens. The cached-read discount rises from 90% to 95%.

Why it matters: Side-by-side intelligence index, cost and token-efficiency figures let readers judge the trade-off GPT-6.1 Sol makes between price and capability.

Sep 29Tuesday

OpenAI News

OpenAI releases GPT-6.1 Sol, nearing GPT-6 Astra at lower price

OpenAI released GPT-6.1 Sol, which approaches GPT-6 Astra on agentic coding, computer use and professional work, at one-fifth the standard input and output token price.

Why it matters: OpenAI cut GPT-6.1 Sol's standard input and output price to one-fifth of Astra's, with several agent task benchmarks landing close to Astra.