Morning edition — Friday 6 November 2026

Google DeepMind released Gemini Omni 1.2 Flash, a video model that generates clips of up to 15 seconds, up from 10 with Omni 1.1, with length set in one-second steps from 3 to 15 seconds. It adds multi-turn editing, so a clip can be changed by follow-up requests instead of being regenerated.

Google cut the price by 30% from Omni 1.1, to $0.07 per second of 720p video. The model is generally available in the Gemini API and on Google Cloud, and in Google Vids and Gemini for business. Google says it holds faces, clothes, packaging text and brand colours more consistently across camera angles when given reference images.

Thinking Machines Lab introduced Inkling 2 Small, the first model of its second generation: a mixture-of-experts model with 594 billion parameters, 17 billion active, that accepts images and lets users set how much effort it spends.

It is available only through the lab's Tinker platform for now, for fine-tuning and sampling, with closed weights. The lab says it intends to release the full weights under an Apache 2.0 licence next week, after external safety testing and red-teaming. Its earlier Inkling models shipped with open weights at launch.

The lab says the model scores 31 on the Artificial Analysis Intelligence Index, far above its earlier models, with the biggest gains in long-running agentic coding. MarkTechPost reports a list price of $0.30 per million input tokens and $1.00 per million output tokens, citing Artificial Analysis.

AWS made bots generally available in Kiro Crew, the agent workspace of its Kiro development environment. Each bot is a named, long-running agent built around one goal, with its own memory, notes and sessions. It can start sessions in parallel, watch them, redirect work that drifts and bring results back to a single chat, and it can run on a schedule, for example to look into overnight build failures.

An AWS spokesperson told The Register that all bots bill together through a developer's existing Kiro plan. Each bot's learned context is kept separate, but all bots share the same workspace, codebase, project files, MCP servers, skills and tools.

The evaluation nonprofit METR published an assessment on Thursday after 11 weeks of high-access work inside Anthropic, from 12 August to 29 October, with internal data on pretraining, reinforcement learning, compute, unreleased models and staff interviews.

Most of the METR staff involved now believe that, as of October 2026, Anthropic is improving its models on well-specified tasks at around twice the rate it did between March 2024 and October 2025, and would not be surprised by anything from 1.3x to 3x. On Anthropic's internal capability index, scores have grown 1.9 times faster since October 2025 (95% interval 1.5x to 2.8x).

METR says it could confidently rule out faster growth in training compute as the cause, and guesses that about 80% of the current speed-up comes from AI used in research. It found no clear bottleneck likely to slow progress over the next 12 months. In a separate review, it said Anthropic's public August risk report understates the acceleration. Co-founder Sam McCandlish told the AP that many of METR's conclusions match his own.

Potential investors in Anthropic's share offering doubt it can hold its planned roadshow next week in time for a November listing, four people familiar with investor discussions told Reuters. The roadshow had been expected to start the week of 9 November, with pricing the following week.

Investors want a more attractive price, and some see the recently discussed $1.5 trillion valuation as the high end, the people said. They also want Anthropic's October revenue, which it left blank in its updated S-1. The Wall Street Journal reported on Thursday evening that IPO investors are sceptical Anthropic can reach that valuation. Anthropic declined to comment to Reuters.

Reuters said Anthropic could also choose to wait if AI shares stay under pressure, for an offering some expect to reach $100 billion.

The neocloud IREN said on Thursday it had delayed some infrastructure deliveries for Microsoft, which it says requested network-design changes, Reuters reported. IREN, which is building 200 megawatts of capacity for Microsoft in Texas, said the delay carries no financial penalties.

It moved the date for reaching its $4 billion annualised revenue target from year-end to January, possibly the end of February, and estimated about $15 billion of capital spending for the rest of its fiscal year. Quarterly revenue was $228 million, ahead of analysts' estimates. The shares closed about 17% lower in Thursday's sell-off.

Bloomberg News reported the same day that Microsoft is tightening oversight of data-centre projects run by neocloud providers including IREN, Nebius and Nscale, Reuters noted.