OpenAI launched GPT-6 Sol and GPT-6 Luna, cutting its mid- and low-cost API prices in half — Sam Altman said both models improve on their GPT-5.6 predecessors in intelligence, alignment, coding, computer use and work output, while Noam Brown highlighted how sharply Luna’s price has fallen. Standard API pricing is now $2 per million input tokens and $10 per million output tokens for Sol, and $0.10/$0.50 for Luna; cached inputs cost $0.20 and $0.01 respectively. Both support a 1.05-million-token context window and 128,000 output tokens. They are available in the API and are rolling out in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu; Free and Go users can try Luna in the desktop app. Tibo Sottiaux also announced a banked usage reset for Plus, Pro and Business accounts. OpenAI’s updated safety material classifies both models as High—but below Critical—in cyber and bio/chemical capability, and reports improved jailbreak resistance alongside some weaker HealthBench results associated with much shorter answers. These are primarily OpenAI-run evaluations, so the launch claims should not be read as uniform real-world gains. OpenAI’s original post · Altman’s capability and price post · Brown on Luna pricing · Sottiaux on pricing and the reset · launch report · API changelog · safety appendix

OpenAI researcher Boaz Barak argues that practical AI-safety measures address more than extinction risk — Barak said he assigns a low probability to literal human extinction from AI, but rejects the idea that discussing it necessarily distracts from nearer harms. His point is that pacing, safety and alignment investment, audits and regulation can also reduce risks such as serious but non-existential incidents, concentration of power, destabilization and gradual loss of control. This is a personal policy position rather than a new OpenAI commitment, but it is a useful statement of where some internal safety thinking overlaps despite different estimates of catastrophic risk. Barak’s post

Anthropic is merging Claude chat and Cowork, while putting Docs, Slides and Design directly inside conversations — Boris Cherny said users will be able to move between quick questions and longer delegated work without choosing a separate surface, with Claude carrying context across the work and continuing after the laptop closes. Claude Docs, Slides and Design can now be created and edited inside a conversation: users can directly change a document, comment on a slide or move a design element, then present, export or share the result. Existing Cowork chats, projects, artifacts, connectors and skills are meant to remain in place. The unified experience is rolling out gradually to Pro and Max over the next few weeks; the three creation tools are in beta on all paid plans. This was announced less than an hour before the scan, so availability will not be immediate or uniform. Cherny on the unified Claude experience · Cherny on Docs, Slides and Design · Claude’s launch thread · Anthropic’s launch post

Anthropic committed to embedded independent evaluators, and OpenAI said it will do the same — Anthropic CEO Dario Amodei proposed “pacing the frontier”: not halting training, but slowing capability gains enough for safeguards and third-party review to catch up. His three-step framework calls for permanent, employee-like access for external evaluators; common safety standards and limits among democratic-country labs; and eventually verifiable global coordination. Anthropic is committing unilaterally to the evaluator step. Sam Altman agreed that the frontier should be paced and said OpenAI will also provide independent evaluators with employee-like access, with details still to come. Amodei’s and Altman’s posts were just outside the strict window—about 27 hours and 24½ hours old respectively—but Anthropic researcher Sholto Douglas endorsed the proposals inside the window, arguing against both an unrestricted race and an absolute pause. These are significant voluntary commitments, but neither company has yet published operational details, timelines or an enforcement mechanism. Amodei’s post · Amodei’s essay · Altman’s response · Douglas’s in-window post

Anthropic released an interactive model for exploring how AI could reshape the US economy by 2030 — Co-founder Jack Clark said the tool varies assumptions about AI capability and diffusion, and is intended to help the public and policymakers prepare rather than predict a single outcome. The model treats occupations as bundles of tasks that AI may augment, automate, leave unchanged or create; its modest, substantial and extreme scenarios all increase total output, while the more transformative cases automate more knowledge work and shift more gains away from labour. Users can set their own assumptions and compare them with a survey of more than 10,000 Americans. These are deliberately simplified scenarios, not forecasts, and they do not cover every possible constraint or outcome. Clark’s post · Clark’s follow-up · Anthropic’s explorer

OpenAI says it has reached its “automated research intern” milestone and released unusually detailed internal adoption data — Researcher Kevin Liu framed the disclosure as evidence for public debate over whether and how to pace model development, and Boaz Barak called AI-assisted AI research one of the most important but least visible inputs into the field’s trajectory. OpenAI defines the milestone narrowly: agents can complete well-defined research tasks under human direction that would take a skilled researcher a few days; this is not a fully autonomous scientist. By mid-August, the median OpenAI researcher was using more than $600 a day of inference at API prices, the 90th-percentile user more than $7,000, and the organization logged 3.1 agent-workdays for every human workday. Code contributions and experiments also rose, but OpenAI cautions that compute increased too, the metrics are preliminary, and more than half of successful four-to-eight-hour tasks still required human intervention. Its target of an automated AI researcher by March 2028 remains a goal, not an achieved capability. Liu’s post · Barak’s post · OpenAI’s report

OpenAI and Anthropic joined a broad call for an urgent surge in AI-enabled cyber defence — Greg Brockman shared an open letter signed by more than 100 organisations, including Anthropic, OpenAI, AWS, Google, Microsoft and Oracle; Sam Altman separately called the moment “critically important” and urged an intense collective response. The letter asks organisations to treat cyber defence as a leadership priority, security vendors to continuously test against frontier capabilities, governments to fund and coordinate protection of essential services, and frontier labs to provide responsible model access, funding, monitoring and verified fixes. Its warning that AI-enabled attacks will become much more widespread and sophisticated “in the coming months” is a signatories’ forecast, not evidence presented in the letter; the concrete news is the unusually broad alignment around defensive action and support for under-resourced critical infrastructure. Brockman’s post · Altman’s post · Open letter and signatories

OpenAI published the first measured results for Jalapeño, its first custom inference chip — Greg Brockman praised the team and linked OpenAI’s new engineering report. OpenAI says that, across GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T on the public InferenceX benchmark, Jalapeño delivered 1.5–1.9 times more peak-throughput work per watt and 1.7–3.6 times lower end-to-end latency than the comparison systems. It also says the chip was taped out in nine months with AI assisting design and verification, and that deployment in OpenAI infrastructure is planned by year-end. These are OpenAI’s first reported measurements, not independent production-scale validation; qualification and broader model testing are still under way. Brockman’s post · OpenAI’s detailed results

OpenAI identified several causes of unexpectedly fast Codex usage depletion and says fixes plus a reset are coming — Tibo Sottiaux said the investigation found inefficiencies when images are used in long sessions with multiple compactions, unusually high p95-and-above usage from Computer History, and a conversation-title feature consuming more usage than intended. He said a dedicated team was preparing fixes for 24 August and that OpenAI would fully reset usage for all paid subscriptions at about 2 pm “PST.” Sottiaux also mentioned a separate, unspecified efficiency approach planned for work next week. These are promised fixes and a reset, not confirmation that they have already landed; the post gives no quantified impact for each cause. Sottiaux’s update · Reset-timing follow-up

OpenAI says Codex has passed 20 million active users and is giving paid users a banked reset — Tibo Sottiaux said Codex crossed 20 million active users during the week and that every Codex and ChatGPT Work user would receive a banked usage reset that can be redeemed later. He separately said OpenAI is still investigating reports of unexpectedly fast usage depletion and has not found a general abnormality so far. One specific pattern did involve unsupported sub2api traffic being flagged by fraud systems; Sottiaux stressed that official clients and open-source clients using Sign in with ChatGPT remain supported. The reset is a one-off credit, not an announced permanent quota increase, and the broader usage investigation remains open. 20-million-user and reset post · Usage-limit clarification