From one frontier model every 17 days to one every 4: AI releases quadrupled in two years
Asked:
“Look up all the AI model releases including open wights ones, and choose the best plot to show how major releases are happening faster.”
223 public frontier language-model releases from 10 major labs, 2018-06-11 through 2026-07-31, extracted from the Top Frontier LLMs Release Tracker. Annual counts rose from 21 (2023) to 55 (2024) to 84 (2025); 61 releases are confirmed open-weight — a lower bound, cross-checked against separate sources. Larger outlined dots mark the 27 models that took the outright benchmark lead.
Open weights confirmedNot confirmed open-weightTook the outright benchmark lead
Cumulative tracked releases over calendar time; each dot is one release. Aligned strip: days since the previous tracked release (log scale) — the floor drops as the curve steepens.
The strict view: gaps between benchmark-leader models
One dot per model that took the outright benchmark lead (temperature2 index). Line: rolling five-record average gap in days.
One tracked release every 17.4 days in 2023 became one every 6.6 days in 2024 and one every 4.3 days in 2025 — a 4× jump in annual volume in two years. mehmetbaykar.com
2026 already holds 55 releases through July 31 — roughly 94 annualized if the seven-month pace holds, though its raw total is not comparable to a full year. mehmetbaykar.com
The rolling five-record gap between benchmark-leader models collapsed from 147.2 days at the end of 2024 to 15.8 at the end of 2025, then rebounded to 28.2 by July 2026 — dramatically faster, but not monotonically. temperature2.com
61 of 223 releases (27%) are confirmed open-weight — weights downloadable, which is not the same as open source; unmatched models are “not confirmed,” not proven proprietary. hidekazu-konishi.com
OpenAI accounts for 63 of the 223 tracked releases, Mistral 32, Anthropic 27, Google 26, DeepSeek 21. mehmetbaykar.com
All 223 tracked releases
#
Date
Model
Developer
Weights
Gap, days
Source
Method: extracted 2026-09-01 from the Top Frontier LLMs Release Tracker (mehmetbaykar.com), which covers public frontier language-model releases from 10 major labs; 223 dated rows, 2018-06-11 to 2026-07-31 (its latest listed release), so not a complete census through September 2026. “AI model” here means public frontier LLMs — not fine-tunes, embedding, image/audio/video models, research checkpoints, or community uploads. “Open weights confirmed” means a same-model, same-date weight-release match in a separate open-weight source search; it is a lower bound and distinct from open source. Benchmark-leader chronology (27 models) comes from temperature2.com’s single index — one operational definition of “major,” not universal truth. Gap values are days between consecutive tracked releases; blank cells were blank in the source.