Nathan Benaich and Air Street Capital have published the ninth State of AI Report. The landing page shows a deck labeled slide 1 of 244, an essay by Benaich, and a citation line: Benaich, N. and Air Street Capital (2026). On X at 07:26 UTC on 8 October 2026, Benaich wrote: “welcome to the 9th annual @stateofai report!” He pointed readers to stateof.ai and to a video he called his editor’s cut. The figures below are from the essay and the pages it links.
Air Street Capital describes itself as a venture capital firm investing in AI-first companies in Europe and the United States. Its team page lists Nathan Benaich as “Solo Member of Investment Staff.” The essay says that since 2013 he has invested in companies building and applying AI, and it names portfolio companies including Wayve and Profluent. That makes the report an investor’s survey of a field he invests in.
What the essay says the frontier looks like
The essay says the frontier is now a three-lab race between Anthropic, OpenAI and Google. It says Anthropic leads on Artificial Analysis’s Intelligence Index, while Google leads on Arena’s ranking of the answers people prefer, “for now.” Those two leads are the essay’s reading of those leaderboards.
The essay also describes a controlled coding study. The link on that sentence goes to an arXiv HTML page, “Stop Comparing LLM Agents Without Disclosing the Harness”. That page’s table, on a 100-task subset of SWE-bench Verified, lists GLM-5.1 at 52.5 under a minimal harness and 65.5 under a full harness, as mean pass@1 over two runs. The essay summarizes the gap as a change in tools, context and feedback, with the model’s weights left unchanged, and calls GLM-5.1 an older model. The 52.5 and 65.5 figures are that paper’s table, as the report cites it. They are a harness comparison, not a claim that the weights improved.
Two lab studies the essay cites
The essay says that according to Anthropic’s internal index, Claude led 26 percent of measured model research and development work in August, up from under 1 percent in February. It says researchers set the tasks and supervise execution, and that Anthropic reports this is speeding development. Anthropic’s own page, “Measurements for understanding the pace of AI development inside frontier labs”, states that as of August 2026 Claude “leads” 26 percent of Anthropic’s AI research and development work, and that Claude is not operating fully autonomously on any measured subset of that work. “Leads,” on that page, means the model can complete most of a task from a high-level prompt while a person supervises. The February “under 1 percent” comparison is the report’s citation of Anthropic’s index. The August 26 percent figure is on Anthropic’s page.
The essay cites an OpenAI study of agent adoption and says it finds the fastest growth among non-developers, including people in legal, sales, recruiting and marketing. That is the report’s summary of OpenAI’s study.
Predictions, graded and new
The essay says the authors grade their own previous calls. The scorecard it describes records two hits, five partial outcomes and three misses. It says they were right that AI labs would lean back into open source to win over the US administration, and that a Chinese lab would top a major task leaderboard. It calls the data-center opposition call a partial hit: the backlash arrived, and the effect on November’s elections was still unresolved when they wrote.
For the next 12 months the essay says there are nine predictions, and it prints three:
- An agent halves its failure rate on new tasks after a month of customer work, without a model upgrade.
- An autonomous AI team beats human-led model research on equal time and compute, setting its own agenda.
- US AI labs officially launch frontier cyberdefense products.
The page says report content is licensed under CC BY 4.0.
Where the numbers come from
The 244-slide count is the landing page’s label. The inner figures belong to the sources the essay cites: Artificial Analysis and Arena for the two leads, Anthropic’s index for the August share, the harness paper for 52.5 and 65.5, and OpenAI’s study for where agent use is growing fastest. None of them is an Air Street measurement.
The Campfire
No commentsNobody has pulled up a log by this one yet. Be the first to say what you make of it.
Held for the desk. It appears after a look.