OpenAI Releases ChatGPT Images 2.5 Image Model

Model Releases & Updates

OpenAI Releases ChatGPT Images 2.5 Image Model

OpenAI releases ChatGPT Images 2.5 image model with up to 50% lower generation latency than Images 2.0, along with improvements in detail, editing precision, reference photo fidelity, and multi-turn consistency.

Source: OpenAI Official (RSS) | Original

---

Product Releases & Updates

Runway Releases Adobe Plugin for Premiere Pro and After Effects

Runway releases Runway Plugins — a new panel that embeds directly into Premiere Pro and After Effects, enabling image and video generation, inpainting, and placement directly on the timeline. Edit Studio re-renders existing clips using Aleph 2 at original duration. Plugin is free to download (macOS and Windows); generation features are available to all paid plans and consume existing credits.

Source: Runway News | Original

OpenAI Rolls Out Astra to All Plus, Pro, Business and Enterprise Users

OpenAI announces Astra is now fully available to Plus, Pro, Business, and Enterprise users across Codex and ChatGPT Work. Users can access it directly, with a live demo available at openai.com/gpt-tv/.

Source: X: OpenAI (@OpenAI) | Original

---

Industry News

NYU Mathematician Accuses OpenAI of Unfair Competition in Millennium Prize Race; Bubeck Denies

NYU math professor Tristan Buckmaster and Anthropic mathematician Levent Alpöge publish preliminary results on three proofs of the Navier-Stokes existence and smoothness Millennium Prize problem. They claim their research progress was leaked to OpenAI, which then used massive compute to follow their unique proof trajectory.

Source: TechCrunch: AI | Original

Mistral Closes €3B Series D at €21B+ Valuation

Mistral announces a €3 billion Series D funding round at a post-money valuation exceeding €21 billion, calling it the largest equity raise in European tech history, just three years after the company's founding.

Source: Mistral AI News | Original

---

Research Papers

OpenAI Announces Navier-Stokes Millennium Problem Solved by Internal AI System

OpenAI announces its internal AI system has produced a solution to the Navier-Stokes existence and smoothness problem, proving that initially smooth fluids can develop singularities in finite time, accompanied by a proof manuscript and Lean formal verification.

Source: OpenAI Official (RSS) | Original

Dwarkesh Patel Research: Pretraining Progress Is Mostly Data

Dwarkesh Patel publishes experimental analysis training 2019–2025 representative model recipes on data corpora up to 1e19 FLOPs. Findings show data improvements deliver 12.0x compute efficiency gains vs. 3.7x for model improvements — data contributing ~3.24x more than architecture.

Source: Dwarkesh Patel: Podcast & Blog (RSS) | Original

---

Tips & Perspectives

Mathematician Buckmaster Publishes Multiple Blowup Results, Discloses OpenAI Communications

Tristan Buckmaster and Levent Alpöge publish three finite-time blowup results covering incompressible porous media equations with smooth forcing.

Source: HN Buzzing (buzzing.cc) | Original

Noam Brown on OpenAI's Agentic Navier-Stokes Proof: Costs Will Drop Rapidly

Noam Brown comments on OpenAI's agentic Navier-Stokes Millennium Prize proof, acknowledging the current cost of millions of dollars but predicting rapid cost declines.

Source: X: Noam Brown (@polynoamial) | Original

GPT-6 Astra操控Blender 完整教程：三种玩法与踩坑实录

A comprehensive guide for controlling Blender with GPT-6 Astra, covering installation, MCP and Computer Use plugin setup, with three practical use cases. Computer Use took ~4 hours to build the Temple of Heaven model but consumed nearly half a $200 Pro subscription. MCP was faster for motorcycle modeling and animation, though complex tasks risked single-run timeouts, solvable by splitting steps or using the CLI to run Python scripts.

Source: WeChat: 数字生命卡兹克 | Original

Sam Altman Responds to Navier-Stokes Proof Dispute with Anthropic Researchers

Sam Altman responds to the controversy around the Euler/Navier-Stokes proof, stating the other party only had Euler results, coordination failed, and they threatened plagiarism accusations.

Source: X: Sam Altman (@sama) | Original

Anthropic: Three Ways to Cut Costs and Boost Performance on Claude Platform

Anthropic's team outlines three methods — optimizing prompt cache hit rates, removing prompt anti-patterns after upgrading to frontier Claude models, and calibrating effort — that reduce costs without sacrificing performance.

Source: X: Claude Devs (@ClaudeDevs) | Original

Tom Tunguz Analyzes Whether OpenAI's 3x AI Productivity Gain Is Just "Machines Don't Sleep"

Tom Tunguz references OpenAI internal data to analyze the 3x research productivity gain: each 8-hour researcher shift equals 3.14 agent workdays, with 4 agents typically running in parallel.

Source: Tomer Tunguz Blog (VC Analysis) | Original

---

【Hacker News Top Stories】

Keywords: AI OR GPT OR LLM OR Claude OR OpenAI OR "machine learning" OR "large language model" OR AGI OR "generative AI" OR AI agent OR RAG OR diffusion OR transformer
Source: hnrss.org | Filtered from Hacker News

1. AI Has Solved One of Math's $1M Millennium Prize Problems

Quanta Magazine reports that AI has solved one of mathematics' million-dollar Millennium Prize Problems.

• HN Thread ->

2. The paradox at the heart of AI and science [video]

A video discussion exploring the deep contradictions between AI and science.

• HN Thread ->

3. Millet: Open source alternative to GPT-4o for document understanding

An open-source GPT-4o alternative focused on document understanding tasks.

• HN Thread ->

4. Show HN: AI Stack Selector Tool

An AI tech stack selection tool to help developers quickly identify the right AI tool combinations.

• HN Thread ->

5. Bullshit arguments about AI (not) replacing jobs (2023)

The author systematically debunks common fallacies about AI "not" replacing jobs.

• HN Thread ->

---

Originally published on WeChat Official Account 「比特财商」.