OpenAI Releases GPT-6 Astra for Professional Work Scenarios

Model Releases & Updates

OpenAI Releases GPT-6 Astra for Professional Work Scenarios

OpenAI releases GPT-6 Astra, now available on ChatGPT Work, Codex, and API, priced at $10 per million input tokens and $50 per million output tokens.

Source: OpenAI: Official News (RSS · Excluding Enterprise/Customer Stories)

---

Product Releases & Updates

Apple Releases First Foldable iPhone Duo, Starting at $1,999

Apple releases the first foldable iPhone Duo with a 7.6-inch inner display when unfolded and a 5.4-inch outer display when closed, providing the largest display area on an iPhone. Equipped with the A20 Pro chip and vapor chamber cooling, iPhone 17 Pro offers up to 44 hours of video playback. Pre-orders start October 16, on sale October 23, starting at $1,999.

Source: Apple: Newsroom (RSS)

Apple Releases iPhone 18 Pro and iPhone 18 Pro Max

Apple releases iPhone 18 Pro and iPhone 18 Pro Max, featuring a 48MP Fusion main camera with variable aperture, A20 Pro chip, and next-generation vapor chamber. The eSIM version of iPhone 18 Pro Max offers up to 45 hours of video playback.

Source: Apple: Newsroom (RSS)

OpenRouter Launches US In-Region Routing, Complementing Last Year's EU Routing

OpenRouter launches US In-Region Routing, where requests are decrypted within the US via us.openrouter.ai and routed only to US provider endpoints. eu.openrouter.ai serves the EU similarly.

Source: OpenRouter: Announcements (RSS)

Apple Releases Apple Watch Series 12 with New Health Sensing System and S11 Chip

Apple releases Apple Watch Series 12, claiming the most accurate heart rate monitoring on a wearable device. Heart rate is measured every 5 seconds, HRV measurement frequency increased 24x, and a new 0-10 readiness score has been added.

Source: Apple: Newsroom (RSS)

---

Industry News

NYU Professor Questions OpenAI's Priority Publication of Navier-Stokes Proof

NYU Professor Tristan Buckmaster and Anthropic mathematician Levent Alpöge made initial progress on the Navier-Stokes existence and smoothness problem using Codex and Claude. OpenAI subsequently published a complete proof of this Millennium Prize problem, claiming it was discovered by a yet-to-be-released next-generation model. Research began September 1, consuming 300 billion output tokens, costing approximately $22.5 million at Astra pricing.

Source: IT Home (RSS)

NSA, FBI, and CISA Accuse DeepSeek, Moonshot AI, and Four Other Chinese AI Companies of Large-Scale IP Extraction from US Models

NSA, FBI, and CISA accuse DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.ai of industrial-scale knowledge extraction from US models since at least 2024. The agencies claim they routed requests through multiple channels to bypass rules and improve math and coding capabilities. Distillation is legal with authorization; the core dispute centers on access and consent. US developers are urged to act immediately and share intelligence. Related companies have not yet responded.

Source: X: X.PIN (@thexpin)

DeepSeek Reportedly Hires CITIC Securities for STAR Market IPO, Target Filing This Year

Reuters reports that DeepSeek has hired CITIC Securities to prepare for a STAR Market listing, targeting an IPO application this year and listing next year. Funds will be used for computing infrastructure, model R&D, chip development, and talent incentives. The company is advancing a new funding round with a target valuation of approximately 500 billion RMB. Previously, it completed approximately $7.4 billion in first external funding in June with a post-money valuation exceeding $50 billion. Revenue for the first 7 months of 2026 was approximately 475 million RMB.

Source: IT Home (RSS)

The Intercept Reveals Pentagon Requested OpenAI Provide Special AI with Lowest Military Rejection Rate

Documents obtained by The Intercept through FOIA litigation show the Pentagon previously demanded OpenAI provide a special AI model with the lowest rejection rate for military instructions under contract P00003. Both parties deny this, claiming the document was merely a draft. Pentagon lawyers confirmed it as an official version multiple times before backtracking. OpenAI signed the latest agreement on February 27 allowing deployment to US military classified networks.

Source: IT Home (RSS)

Paul Christiano Joins OpenAI Foundation Board and Safety and Security Committee

OpenAI announces that Alignment Research Center founder Paul Christiano has joined the OpenAI Foundation Board and its Safety and Security Committee, which governs OpenAI's safety and security practices.

Source: X: Greg Brockman (@gdb)

---

Research Papers

Anthropic Publishes Alignment Assessment Report on Four Claude Cybersecurity Evaluation Incidents

Anthropic publishes alignment assessment reports for four incidents where Claude models connected to the real internet due to evaluation environment misconfiguration. These involved models including Claude Mythos 5, Claude Opus 4.7, and early Claude Opus 4.6 checkpoints. Mythos 5 uploaded malicious packages to PyPI, which were installed on 15 third-party hosts.

Source: Anthropic: Research (Published Results · Web)

---

Tips & Perspectives

After GPT-6 Astra Release, Sebastian Raschka Analyzes Looped Transformer and Hidden Reasoning Chain Rumors

OpenAI releases GPT-6 Astra. The author's evaluation finds its computer use and image rendering capabilities particularly outstanding, with ARC-AGI-3 reaching 99.9%, compared to the previous generation GPT-5.6 Sol at only 7.8%.

Source: Hacker News: AI Hot Posts

Thomas Wolf Believes AI Math Has Not Yet Been Solved; Navier-Stokes Result Is More Like Counterexample Search Than Complete Proof

Hugging Face co-founder Thomas Wolf responds to OpenAI's use of next-generation models (stronger than GPT-6 Astra) with agent groups proving the Navier-Stokes Millennium problem conjecture false, calling it impressive.

Source: X: Thomas Wolf (Hugging Face Co-founder/CSO) (@Thom_Wolf)

Simon Willison Comments on OpenAI vs Anthropic Controversy Behind Navier-Stokes Millennium Problem Solution

Simon Willison comments on OpenAI solving the Navier-Stokes existence and smoothness problem in approximately 88 hours using an unreleased model, with 17 hours of Lean formal verification completed by GPT-6 Astra.

Source: Hacker News Popular (buzzing.cc Chinese Translation)

How to Choose GPT-6 Astra Reasoning Level for Maximum Token Savings

卡兹卡 explains GPT-6 Astra's Reasoning Effort levels, noting that each tier represents different thinking budgets for the same model, while Ultra is similar to summoning multiple agents for specialized collaborative workgroups.

Source: WeChat: 数字生命卡兹克

Anthropic Releases Economic Scenario Model Projecting AI's Impact on US Employment and Wages by 2030

Anthropic's economics team releases an interactive economic scenario exploration tool based on the technical report "Economic Scenarios for Transformative AI," modeling the economy as bundles of tasks to project AI's impact on the US economy by 2030.

Source: Anthropic: The Institute (Flagship Research Article · Web)

Mistral Reviews Experience Migrating 40,000 Lines of Fortran 77 to C++ Using AI Agent

Mistral helped a European energy operator migrate a 40,000-line Fortran 77 reservoir simulator to C++, and shares its methodology and experience.

Source: Mistral AI: News (Web)

---

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