Plotting AI model release cadence: two labs are accelerating, three aren't

Analysis · June 2026

Plotting AI model release cadence: two labs are accelerating, three aren't

Plotting frontier model release cadence, with methodology

SwiftAlerts · June 20, 2026

Ethan Mollick made an offhand observation this month: if AI self-improvement is real, even weakly, then the labs that have it should ship faster over time, and the ones that don't should fall behind. He claimed this was already visible at Anthropic and OpenAI but nowhere else. I wanted to check whether the release data actually supports that, so I plotted it.

If AI self-improvement, even in a very limited way, is possible, the cadence of shipping both AI products, harnesses, and models should go up. This appears to be happening at Anthropic and OpenAI, but not for any other labs, including those that seemed to be catching up last year.<br>Ethan Mollick, June 19, 2026 [1]

The claim is falsifiable, which is rare for AI-progress takes, so it's worth testing against data rather than vibes. Here's the cumulative count of major frontier model releases per lab since Q1 2023.

Anthropic (13)<br>OpenAI (11)<br>Google (8)<br>Meta (7)<br>DeepSeek (5)

Cumulative releases by Q2 2026: Anthropic 13, OpenAI 11, Google 8, Meta 7, DeepSeek 5.

Cumulative count of major frontier model releases per lab, Q1 2023 to Q2 2026. Slope is cadence. Sources in notes [2].

Methodology & caveats

What counts as a release: a distinct, publicly available frontier or flagship model or major version bump (GPT-4, GPT-4o, o1, o3, GPT-5, GPT-5.5; Claude 1 through Opus 4.8; Gemini 1 through 3.5; Llama 1 through 4; DeepSeek V2 through V4 and R1). Point releases and minor checkpoints are excluded to avoid rewarding version-number inflation. Reasonable people will draw this line differently, and the exact counts shift a little if you do.

Cadence is a proxy, not proof. A steeper slope is consistent with recursive self-improvement, but also with more funding, better management, or simply a decision to ship more often. This chart shows the pattern Mollick described exists in the data. It does not prove the causal mechanism.

Cumulative counts can mislead. A rising cumulative line only means releases are still happening. The signal worth caring about is the second derivative, whether the slope itself is steepening. That's the second chart.

The thing to look at is which lines are bending. Anthropic and OpenAI don't just have the steepest slopes, their slopes increase toward the right. Google sat nearly flat through 2025, then sprinted in Q2 2026. Meta plateaued after Llama 4 in April 2025 and hasn't shipped a frontier model since. DeepSeek runs a steady quarterly cadence without accelerating.

To isolate acceleration, here's the annualized release rate, a trailing four-quarter window. On this view a flat horizontal line means constant cadence; an upward-bending line means accelerating cadence.

Anthropic<br>OpenAI<br>Google<br>Meta<br>DeepSeek

Annualized rate Q2 2026: Anthropic 6, OpenAI 5, Google 4, Meta 0, DeepSeek 2.

Annualized release rate, trailing four-quarter window. Flat line = linear cadence. Upward bend = accelerating cadence.

Two labs bend up. Three don't. Anthropic roughly tripled its annualized rate over the window; OpenAI more than doubled. Google held flat until a 2026 catch-up; Meta is in decline.

The recursion argument

There's a deflationary reading where this is just spending and headcount, and the cadence gap won't compound. The argument that it does compound rests on a specific loop: the labs use their own products to build their successors. Anthropic engineers use Claude Code to write training and eval infrastructure for the next Claude. OpenAI uses Codex on Codex. Each release improves the harness that produces the next release, so the next one ships sooner and better.

Note what this is and isn't. The deployed model is frozen between versions, so there's no online learning happening inside the weights. The recursion is at the level of the organization, not the model. Call it offline RSI: the loop closes across release cycles rather than within a forward pass. That's a much weaker claim than "self-improving AI," and it's the one the chart is actually consistent with.

Two other things landed in the same window that the recursion reading predicts. First, compute efficiency: Tri Dao's FlashAttention-4 hit 71% utilization on NVIDIA B200 in March 2026 [3], and Mamba-3, from the same group, was explicitly designed inference-first rather than training-first [4]. Cheaper training and inference per cycle means more cycles per quarter. Second, talent concentration: in the week of June 19, Noam Shazeer (Transformer co-author) joined OpenAI to lead architecture research, and John Jumper (AlphaFold, 2024 Nobel) left Google DeepMind for Anthropic [5]. Talent is flowing toward the labs already shipping fastest.

What would falsify this

The honest failure modes, since the whole point was...