Google's Gemini 4 Carbon tops AI index as Gartner says open models can do most agent work

Gemini 4 Carbon (high) scores 59 on the Intelligence Index V5 benchmark suite, Tom's Hardware reported on 3 November. That is one point ahead of Claude Opus 5.5 (max with fallback). It is two points clear of OpenAI's GPT-6 Astra (max).

The model also tops Artificial Analysis' new AA-Omniscience v1.1 test, which tracks hallucination and recall. Google is opening access to cyber defenders first through its Fairwind program. Wider availability comes later.

Read the fine print and the win shrinks. Even the best score on that omniscience test was 54 on a -100 to +100 scale, Tom's Hardware noted. That suggests frontier models in 2026 remain far from consistently reliable at recalling information. Gemini 4 Carbon costs $7.76 per task on the Intelligence Index. That is slightly cheaper than Gemini 3.8 Flash (high) and roughly half the price of Claude Fable 5.1 at $12.99. It is still about 13 times as expensive per task as the open-weight DeepSeek V4.1 Flash 1017, which scores 32.

That gap is the backdrop for a more consequential claim from the same week.

Gartner puts a number on the open-weight case

Analyst firm Gartner argues that nine in 10 agent workflow steps are routine. Open-weight models handle them at parity for a quarter of the price of the cheapest closed model, Gartner says. The Register reported the document, titled "Open-Weight Models Will Run 90% of Enterprise Agentic Workflow Steps," on 3 November.

The authors are senior director Mark Petty and group vice president Raymond Paquet. They wrote that open-weight models "can handle hundreds of these routine steps for every one that needs frontier reasoning." Against a frontier model run on every step, they put the cost at as little as a twentieth.

"Current open models can be more verbose and retry more often, so the saving per task against a frontier model is closer to five or six times."

Gartner's planning assumption is that by 2030, 90% of enterprise operational agent workflow steps will execute on open-weight models. The analysts frame the money as secondary to control. Routine steps sent to a frontier API move workflow history outside a company's perimeter, they argue. The cost saving is a bonus.

Their paper also points at harnesses, the software that routes work to the right model, as where value sits in AI products. In that reading, frontier models are capital expense for research and development. Open-weight models become the operating expense.

Robotics money keeps moving

On 2 November, more than 84% of votes cast at a special shareholder meeting approved Churchill Capital Corp XI's business combination with Agility Robotics, The Robot Report reported on 3 November. The SPAC is expected to close on 4 November. The combined company is due to trade on Nasdaq under AGLT from 5 November.

The deal is now expected to provide Agility with roughly $465 million in gross capital before expenses. That is down from the more than $620 million anticipated when the merger was announced in June. It breaks down as about $164 million from Churchill's trust account, about $201 million from a PIPE led by Foxconn and $100 million in equity Agility received after the initial announcement. The original transaction valued Agility at a $2.5 billion pre-money equity value.

Agility says its Digit robots are already operating in nine customer environments. Logistics and manufacturing are the main markets. It has claimed more than $300 million in committed multi-year orders for Digit 5. The Robot Report cautioned that the figure is not current revenue and depends on contractual milestones.

Tutor Intelligence, meanwhile, said on 2 November that it raised $75 million in Series B funding led by Andreessen Horowitz. The Watertown, Mass. company spun out of MIT CSAIL in 2021. It pays its human "tutors" to supervise robots in production and uses the resulting data to train models. It says twice as many of its robots are working in almost four times as many facilities as a year ago. It says it is on track for more than 1,000 robots under contract by the end of 2027.

The data problem underneath all this is getting attention too. VicOne identified 28 zero-day findings in less than two days using AI-assisted research across software supporting the physical AI ecosystem, The Robot Report reported on 1 November. Of the findings, 75% involved development, training, simulation or validation tools. None of that is an attack. It is a reminder that the tooling feeding robot models is itself software with a supply chain.