AI news, roughly super. A daily briefing from what the AI YouTube world actually said.

Thursday, September 10, 2026

Coverage: 94 videos reviewed (0 partial or status unknown); videos under 45 s were not reviewed. Dates are UTC upload dates. Vendor claims are labelled as such.

New today

OpenAI reports its model found a finite-time blow-up solution to Navier-Stokes
OpenAI reported that an AI model found a very likely finite-time blow-up solution to the Navier-Stokes existence and smoothness problem, according to Two Minute Papers and Matthew Berman, who relayed the claim on Sept. 10, 2026. Two Minute Papers said the work took about 3.5 days; Berman said OpenAI's report put it at five days and that an internal model more capable than GPT-6 Astra was used. A Fireworks AI presenter, who said he was unsure of details, recalled OpenAI claiming tens of thousands of agents, more than $10 million and 88 hours. None of the speakers verified the proof, and the Two Minute Papers host said he is not an expert on the problem. Two Minute Papers also read an OpenAI reply saying it cannot rule out that two outside scientists' chat data helped improve its models.

GPT-6 Astra is rolling out on paid ChatGPT plans, the API and AWS, speakers say
Nate B Jones said on Sept. 10, 2026 that GPT-6 Astra is rolling out on paid ChatGPT plans, the API and AWS; he did not verify availability. Leon van Zyl reported that at the time of his recording normal ChatGPT chat sessions did not yet offer GPT-6, so he selected Astra in the desktop app's Work mode at high to extra high reasoning. Availability is as of each recording date and is not an OpenAI announcement.

DeepSeek released V4.1 Flash with MIT-licensed weights and native image input
DeepSeek released V4.1 Flash, according to reviewers AI Code King and Bijan Bowen, who relayed the company's technical report and release page on Sept. 10, 2026. They described a 552 billion-parameter mixture-of-experts backbone plus 196 billion engram-memory parameters, about 8 billion active on input and 16 billion on output, native image input and Hugging Face weights under an MIT license. DeepSeek reported 74.2 on DeepSWE 1.1 versus 62.7 for V4 Pro at max effort, and charts showing it comparable to Kimi K3; the reviewers did not verify these. Bowen speculated, with a caveat, that the engram parameters could be offloaded to CPU memory or SSD.

GitHub and Microsoft Research report Hydra Fusion model routing cuts cost 36-67% versus Opus 5
GitHub said its Hydra Fusion research preview routes tasks to a single model, a cheap-then-escalate cascade or a draft-and-critique pair, and is an experimental option in the Copilot CLI. Microsoft Research's Ashna Garg reported, from vendor-run offline evals, 67% lower cost than Opus 5 on Terminal Bench 2.1, similar quality at 36% lower cost on DeepSWE and similar quality at 65% lower cost on an internal checkpoint benchmark. A four-task live demo came in 42% below Opus 5 and 47% below Fable 5.1 on cost, per the presenter. No run counts or raw scores were shown.

OpenAI launches Agents API with hosted Codex harness, MCP tools and multi-agent delegation
OpenAI's video presented an Agents API that runs a hosted Codex harness with sessions, orchestration and context management, tools via MCP, runbooks as skills and bring-your-own sandbox. It also lists programmatic tool calling, multi-agent delegation and compaction. Pricing, limits and availability were not stated, and token savings were not quantified.

Continuing stories

Also notable

Models & learning