Sunday: a conjecture solved, a model family launched, and a regulatory first
A week that tested what frontier AI can do — and what governments will let it do. From a claimed proof of a 50-year-old math problem to the first-ever government-forced suspension and reinstatement of a frontier model, the throughline is the same: capability is outrunning the frameworks built to contain it.
GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture
A claimed proof of the Cycle Double Cover Conjecture — open since the 1970s — would be the clearest demonstration yet of AI as a genuine mathematical collaborator, not just a tool.
Claude Fable 5 and Mythos 5 reinstated after U.S. export control suspension and safety guardrail negotiations
The first government-forced suspension and negotiated reinstatement of a frontier model sets a template — and a warning — for how capability reviews could reshape every future top-tier release.
OpenAI releases GPT-5.6 model family with Sol, Terra, and Luna tiers
GPT-5.6's three-tier family and new API primitives — persisted reasoning, programmatic tool calling, multi-agent orchestration — signal where the platform layer is heading for developers.
Meta Superintelligence Labs releases Muse Spark 1.1, a multimodal agentic reasoning model with Meta Model API
Muse Spark 1.1's public API debut gives developers a new frontier-tier multimodal agent to benchmark against — and marks Meta Superintelligence Labs as a serious platform contender.
Get the next edition in your inbox
Midweek and Sunday. Primary-sourced AI news paired with the guides that explain it.