This refresh widens the tracker's lens. Edition 03's original discovery method — GitHub-API-only, no roundups or aggregators — still works for measuring open-source star velocity, but it structurally cannot see funding rounds, foundation-model drops, or product launches from companies that don't live on GitHub. Covering those means drawing on secondary sources, so this refresh does that openly: every item below is tagged Verified (confirmed via GitHub, a company's own press/changelog, or comparable first-party record) or reported (resting on a single aggregator or secondary source no lab or company has directly confirmed) rather than presented as uniformly solid.
AI Tracker · Edition 03 · Refresh · 9 October 2026
OpenAI Shipped GPT-6 and Agents That Keep Working After You Log Off. Google Kept Its Best Model for Cyber Defenders.
Revised · v5.0.0 · Edition 03 refresh · 17 new movers this window
Consistency & methodology notes
A small verified sample was checked first (Ruflo, Graphify, Kolibri, HappyRobot), each fact checked against its primary source, then scaled to the full board across all five categories. Items are labelled Verified where a figure came from GitHub itself or reputable first-party press, and reported where it rests on a single aggregator or secondary source a lab or company has not directly confirmed.
What changed this refresh (as of 9 October): 17 new movers added. The window was dominated by OpenAI’s GPT-6 rollout and its 29 September DevDay (Agents API, Dots, Codex Cloud), Google’s Gemini agent for work and its restricted Gemini 4 Argon, and Meta’s consumer Muse agent. Six existing items were updated, four carried over unchanged, and thirteen were retired: Konecta Kolibri, Sapiom, Convex, HappyRobot, Drata, the July Codex price cut, a July Anthropic model release, DeepSeek-V4-Flash, Kimi K3, FLUX 3 Video, Muse Glimmer, Google Calling Agents and open-connector are off the board because they predate the window, were superseded, or had no new development.
- GitHub star counts were not refreshed. Ruflo (67.6k stars / 8.1k forks) and Graphify (16 stars / 4 forks) are still the 11 August 2026 snapshot, because GitHub’s API could not be read from this refresh’s tools.
- Name-collision watch — “Graphify”: the reference example is a small repo (rhanka/graphify). Several SEO blogs claim “85,000+ stars” for “Graphify” — that figure is unverified and does not match any live repo.
- Name-collision watch — “Kolibri”: Aleph Alpha released an open-weight model called Kolibri-1 on 3 October 2026. It is unrelated to Konecta’s enterprise agent platform “Kolibri” (June 2026) or to GitHub repos spelled “colibri”.
- Dates that differ between sources are shown as ranges, not resolved: Cursor Origin (changelog 17 Aug, an older listing 13 Aug, TechCrunch 18 Aug), Qwen3.8-Flash-Next (repo 24 Aug, press 26–28 Aug), UiPath voice agents (14 or 17 Sep), and Qwen3.8-Max open weights (repo 8–12 Aug, CNBC 17 Aug).
- Verified on this refresh means a first-party page, a first-party repository, or reputable press (TechCrunch, CNBC, Bloomberg) was found by its own URL; most pages could not be opened directly, so benchmark and pricing figures are vendor-reported. Gemini Spark’s $99.99/mo price is still single-source.
- Funding: Temporal’s round is confirmed on its own blog (Bloomberg had reported talks at about $500M); Nous Research and AIR rest on TechCrunch reports and are marked “reported” until a company announcement is linked. Larger rounds outside agent infrastructure (Instinct, $1B; Manus, over $500M) are not on the board. One aggregator’s “Composio $25M Series A, 14 Sep 2026” is a recycled 2025 round and was ignored.
- Freshness bias: the board prioritises roughly the last eight weeks. Carryovers say what, if anything, changed this window.
Sources. Star/fork counts are read directly from the GitHub API. Launch, funding and model-release items are sourced per-item as linked in each card above and in the notes; see each card's “Source” link for the specific citation. Prior editions are linked from the AI Tracker hub.
About this article: Researched, written and edited by Umashankar Triplicane Dwarakanathan, with AI research assistance; every figure is meant to trace to the primary source cited. See the Editorial Policy for how sourcing, AI use and corrections work.
