OpenAI spent the week on two fronts: a new model preview and its first custom chip. At the same time, the White House pushed for tighter rollout controls, while Amazon, SpaceX, and Agility Robotics moved money into compute and hardware.
Model releases & updates
- GPT-5.6 Sol was previewed by OpenAI with stronger coding, science, and cyber performance, plus added safety work. That puts more pressure on rivals to match capability without loosening controls.
- OpenAI’s Jalapeño chip is its first custom inference chip, built with Broadcom. It gives OpenAI more control over serving costs and latency instead of relying entirely on off-the-shelf silicon.
- Asian startups launched Mythos-like models while Anthropic’s export restrictions continue. That suggests model development is spreading even where access to leading U.S. systems is limited.
- Anthropic’s Mythos was cleared for select U.S. companies and government agencies under controls. Access is still being managed model by model, not opened broadly.
Money & moves
- Amazon committed another $13 billion to AI infrastructure in India. That is a direct bet on local compute, data, and cloud demand in one of the biggest growth markets outside the U.S.
- SpaceX signed a compute deal with Reflection AI worth up to $6.3 billion. The size of the contract shows how expensive frontier-model training and inference capacity has become.
- OpenAI has not held pre-IPO investor meetings or set a timeline, according to sources. That keeps a major liquidity event off the table for now, even as investor attention stays high.
- Agility Robotics plans to go public through a $2.5 billion SPAC deal. Robotics capital formation is still alive, but it is leaning on public-market shortcuts.
Policy & governance
- The White House asked OpenAI to slow the release of GPT-5.6 over safety concerns. Government review is now affecting model launch timing, not just post-release rules.
- OpenAI will delay GPT-5.6 after a Trump administration request over security concerns. That makes the release schedule a policy issue, not only a product decision.
- OpenAI is limiting new models to “trusted partners” at the U.S. government’s request. If this sticks, model access could become more gated for enterprise buyers.
- Anthropic’s Mythos 5 reportedly returned after negotiations with the Trump administration. The result is a reminder that export controls can change which labs reach which customers.
Research
- The Capability Frontier argues benchmarks miss 82% of model performance. That is a direct warning that leaderboard gains can understate real-world failures and strengths.
How this digest is made
This roundup is generated by Letaido, an AI agent that runs the whole pipeline automatically. Each week it pulls the latest posts from a curated set of AI-industry sources — major tech-press outlets, the AI labs’ own blogs, arXiv, and live web-event streams from Firehose taps — then uses AI to score every item for relevance and importance, drops near-duplicates, and ranks what made the cut. This issue was drawn from 672 stories across 10 sources; 18 made the final digest. A human editor reviews each issue after it publishes and their feedback tunes future editions. Sources are linked inline so you can read the primary reporting yourself.
