AI Model Releases in 2026: Which Labs Ship Most
50 named AI models launched in 12 weeks, about one every 1.7 days. 14 labs shipped them; none of them can count on a quiet month.
The signal
- LaunchesGrok 4.7 (21 Sept 2026) shipped at $2/$6 per million tokens, the same list price as Grok 4.6, xAI's launch number 5 since July.
- LaunchesQwen3.8-Omni-Flash (20 Sept 2026, per launch trackers) is Alibaba's launch number 6 since July, one of 2 Qwen models listed that day.
- ArenaThe 25 Sept 2026 Arena text board ranks 409 models from 53 organisations.
14 labs launched 50 named AI models from 6 Jul 2026 to 27 Sept 2026. Alibaba's Qwen team led with 7; Chinese labs made 19.
Named model launches per lab, 6 Jul 2026 – 27 Sept 2026. One bar per lab, longest first. Emerald = labs based in China.
Counts follow our launch feed row by row. A post that names several variants counts once (4 rows, such as "GPT-6 Sol / GPT-6 Luna"); sibling models listed separately, such as Gemini 3.8 Flash TTS and Flash-Lite TTS, count separately. Rows from 6 Jul 2026 to 22 Sep come from launch trackers and are a floor; the 6 rows from 23 Sep come from a sweep of lab posts. Some dates differ by one to three days between trackers and lab posts.
Show the data (14 rows)
| Lab | Launches |
|---|---|
| Alibaba (Qwen) | 7 |
| 6 | |
| Meta | 6 |
| OpenAI | 6 |
| xAI | 5 |
| Anthropic | 3 |
| ByteDance | 3 |
| NVIDIA | 3 |
| Tencent | 3 |
| DeepSeek | 2 |
| Moonshot (Kimi) | 2 |
| Z.ai | 2 |
| Black Forest Labs | 1 |
| Runway | 1 |
The Mode read
xAI shipped Grok 4.7 on 21 Sept 2026 and Alibaba listed 2 new Qwen models the day before. Each was one more row in a busy feed. How many models do the labs now release, and who releases the most?
Our launch feed counts 50 named models from 14 labs in 83 days, one every 1.7 days. Alibaba's Qwen team is the busiest, with 7, one every 12 days or so. Google, Meta and OpenAI follow with 6 each. The top five labs made 30 of the 50. Labs based in China, 6 of them, made 19.
When a capable new model lands every couple of days, mostly from the US and China and from 14 labs, no single release is scarce. The supply side looks like a commodity market with many sellers. None of these labs sells into most industries, languages or regions directly; they sell an API and wait. The value sits with whoever turns this steady flood into one product for a market the labs cannot reach on their own.
14 labs supply the models. None of them owns your customer.
What we would back: Founders who turn the steady stream of US and Chinese models into one product for a market the labs will not sell into directly: a regulated industry, a language, a region.