OpenAI formed an independent advisory group on mathematics and artificial intelligence hosted by the Institute for Advanced Study in Princeton to provide mathematicians input on its math research. OpenAI said its internal model has resolved more than 100 additional open problems. The group will offer advisory and release-coordination input without decision-making power or authority to change OpenAI’s pace.
MiMo launched MiMo-V2.6, Xiaomi’s open omnimodal model family for long-horizon agent work. It supports 1M context in its Pro and Flash variants. Xiaomi also released a technical report plus RL environments and training code for the public post-training run.
TypeSafe launched Jev, a “System One” text-only model meant for software decision-making rather than chat or writing. Jev’s input tokens cost $0.042 per million, and most calls are reported to finish in about 100 milliseconds. As a result, developers can embed Jev as a typed decision module inside applications with calibrated probabilities and configurable autonomy thresholds, changing how AI is integrated into products.
AI agents generate text for decisions that can be made without words, adding token and compute overhead. The article cites OpenAI researchers spending $7,000 a day running agent workloads, and describes Jared Palmer releasing Kev models (0.8B, 4B, 9B) that do a single prefill-only forward pass with pointer-head probabilities instead of autoregressive decoding. Decision-making is shifted to probability scoring over a provided candidate list, with open-ended reasoning left to larger generative models and multi-decision batching supported.
SpaceXAI released Grok 4.7, a coding agent model trained to handle long, hours-long runs by focusing reinforcement learning on harder tasks and improving self-verification and long-context management. Grok 4.7 scored 38.0% on Terminal-Bench 4.0, up from 20.3% for Grok 4.6. The model’s long-horizon performance improved substantially, but it still fails most of the time in long-running codebase evaluations, so reliability remains a limiting factor.
RetroChimera, a retrosynthesis prediction framework, was published in Nature and released as open-source to automatically propose synthesis routes for small molecules. In blind tests, PhD-level chemists preferred its individual reaction predictions over the preceding models and recorded literature reactions. The release enables researchers to use and improve an ensemble model that combines two approaches to rank routes better and scale synthesis planning beyond manual processes.
Grok 4.7 was launched as a more capable model for coding and knowledge work. It is served at the same price and speed as Grok 4.6. The update focuses on longer performance on difficult tasks, more careful self-checking, and improved safeguards.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
