Andreessen Said AGI Crossed in February. The Model Release Dates Say Otherwise.
Marc Andreessen went on Joe Rogan's podcast on May 19 and declared that the AGI threshold was crossed roughly three months earlier, around February 2026. "I actually think we crossed that about three months ago," he said. The claim got attention. The models he cited as proof tell a different story from the date he assigned to the crossing.
The timing matters because Andreessen's firm, Andreessen Horowitz, manages roughly $90 billion in assets under management as of early 2026 and has significant AI sector exposure through its portfolio, including disclosed stakes in OpenAI and Anthropic. His declaration that the most consequential technological threshold has been crossed is also a statement about where capital his firm has already deployed stands to gain. That alignment of the person drawing the line with those who stand to benefit is the structural problem the story is about.
Andreessen named four models as evidence: GPT-5.5, Claude 4.6, Gemini 3.0, and Grok 4.3. Two were released after the threshold he set. GPT-5.5 shipped April 23, 2026, according to CNBC's coverage of the announcement, and Grok 4.3 shipped May 6, both after February. Only Claude Opus 4.6, released February 5 by Anthropic, fits his stated timeline. Google DeepMind's most recent publicly confirmed frontier releases as of May 2026 are Gemini 3.1 Pro and 3.5 Flash. As of May 2026, no version numbered 3.0 appears in any verified release announcement from Google or DeepMind, and the public record shows no 3.0 release. The chronological foundation of Andreessen's claim does not hold against the evidence he cited.
AI commentator Ole Lehmann posted a fact-check thread the same day the Rogan episode dropped. It pulled 694 likes. The contradiction was noticed and noted, though Andreessen did not update his claim.
On the Rogan episode, Andreessen said that 99 percent of the time, the answer from the most advanced AI models is better than what he would get from talking to almost any expert he has access to (a paraphrase of his phrasing, per OfficeChai's coverage of the interview, not a direct transcript). That claim, if it were a controlled benchmark, would be significant. As personal experience reported on a three-hour podcast, it is not a controlled measurement.
There is a pattern here worth noting. Bill Gates wrote his famous "internet tidal wave" memo in 1995, after Microsoft had already placed its bets. Jeff Bezos declared AWS a strategic necessity for Amazon in the early 2000s, before the financial case was clear: Amazon's position in cloud is now its most valuable franchise. When the person with financial interests in the outcome is also the person drawing the line society uses to define success, the line tends to get drawn where the investment already lives. Andreessen is not the first to do this. He is the latest.
AI researchers who have spent careers studying the field are less definitive. Fei-Fei Li, a co-director of Stanford's Human-Centered AI Institute, has called AGI more of a marketing term than a scientific one, noting no consensus exists on a single definition, per OfficeChai's coverage of the interview. Demis Hassabis at Google DeepMind has argued that current systems still fall short because of inconsistencies and easily exploited weaknesses, gaps that rule out genuine general intelligence in any rigorous sense.
Andreessen did not cite a benchmark, a test, or a published metric. He pointed to a cluster of recent model releases and declared a threshold crossed. Whether that threshold was crossed in February, April, or May matters less than the fact that the specific evidence he offered for a February crossing does not hold. The definitional question, what AGI means, how to measure it, when it would matter, is a separate and still-unresolved debate. What Lehmann caught is concrete: the models Andreessen cited as proof are not consistent with the date he assigned to the crossing.
Every board and strategy team that has been waiting for permission to bet big on AGI-level capabilities just received it from someone whose firm has significant AI sector exposure. Major technology companies have disclosed nine-figure AI infrastructure commitments in the past twelve months, and a public milestone declaration from a prominent venture investor gives board members a reference point they can cite in fiduciary deliberations. That cover is exactly what makes the declaration actionable in a way a research paper is not. The loudest microphone just drew the line, and it landed exactly where the investment already lives.
What comes next is a test of whether the declaration changes behavior. If infrastructure spending accelerates, if AI headcount allocations shift, if the next earnings call features more AGI-language than the last — that will be the real signal, independent of what the model release dates actually say.