A Timeline of AI Model Releases in 2026 (Updated Through July)
New frontier models are shipping every few weeks in 2026. Here's a running timeline of the major releases from OpenAI, Anthropic, Google, xAI, and the leading open-weight labs, so you don't have to track it yourself.
- 012026 has seen a new frontier or near-frontier model release roughly every two to three weeks across the major labs.
- 02Anthropic, OpenAI, and Google have all shipped multiple model generations so far this year, not just a single annual update.
- 03Open-weight labs, especially DeepSeek, Qwen, GLM, and Kimi, are releasing on a similar or faster cadence than the proprietary labs.
- 04This page will be updated as new models ship; treat any single snapshot as a point-in-time reference rather than a permanent ranking.
If keeping up with AI model releases feels like a full-time job, that's because the pace genuinely has picked up. This post is a running log of the major model releases in 2026, meant to complement our guide to understanding the different AI models with the specific dates behind each family's current lineup.
TL;DR: Every major lab has shipped multiple model versions in 2026 rather than a single annual release. Anthropic moved through several Claude generations and introduced a new Mythos tier above Opus. OpenAI shipped the GPT-5.4, 5.5, and 5.6 lines. Google pushed several Gemini 3.x updates. xAI released Grok 4.20 and Grok 4.5. Open-weight labs like DeepSeek, Qwen, GLM, and Kimi kept pace with frequent releases of their own.
Early 2026: the year starts fast
The first few months of 2026 set the tone for the rest of the year, with dense, overlapping release windows across labs. Anthropic shipped Claude Sonnet 4.6 and Opus 4.6 in February, followed by Claude Opus 4.7 in mid-April, aimed at complex reasoning and long-running agent workflows. Google released Gemini 3.1 Pro and Flash-Lite in February and March, then Gemini 3.1 Ultra with native multimodal reasoning shortly after. OpenAI released GPT-5.4 in three variants (Standard, Thinking, and Pro) on March 5, adding native computer-use capability. xAI shipped Grok 4.20 Beta 2 on March 3, with real-time web and X data access built in.
On the open-weight side, Alibaba's Qwen 3.5 and Z.AI's GLM-5 both landed in this window, each closing more of the gap with proprietary mid-tier models, and Google released Gemma 4 under an open license on April 2.
Spring 2026: the open-weight labs accelerate
April and May saw open-weight labs pick up pace. DeepSeek shipped DeepSeek-V4-Pro and DeepSeek-V4-Flash on April 24, later making a major pricing discount permanent in May. OpenAI followed with GPT-5.5 and GPT-5.5-Pro on April 23. Moonshot AI released Kimi K2.6, with K2.7 Code following in June. Mistral shipped Mistral Medium 3.5 in late April. In May, Alibaba previewed Qwen 3.7-Max, Google released Gemini 3.5 Flash, and xAI put a Grok Build CLI into beta.
Summer 2026: a new Claude generation and rapid-fire updates
June brought one of the busiest stretches of the year. Anthropic released Claude Opus 4.8 in late May, followed by the first models in the new Claude 5 generation: Claude Fable 5 on June 9, introducing a new Mythos-class tier above Opus (with Fable 5 generally available and Mythos 5 limited to approved organizations), and Claude Sonnet 5 on June 30. Z.AI released GLM-5.2 on June 16. OpenAI shipped the three-track GPT-5.6 line (Sol, Luna, and Terra) in late June, with a 1-million-token context window.
Into July, releases kept coming at a similar pace: xAI's Grok 4.5 on July 8, Moonshot AI's Kimi K3 on July 16, Google's Gemini 3.5 Flash Cyber and Gemini 3.6 Flash on July 21, and Anthropic's Claude Opus 5 on July 24.
What this pace means for you
A few practical takeaways follow from this cadence:
- Don't over-invest in a single model version. Whatever you build today will likely have a newer, better, or cheaper option available within a matter of weeks.
- Watch pricing pages, not just release announcements. Labs frequently adjust pricing after a release, sometimes making a previously expensive tier permanently cheaper, as DeepSeek did in May.
- Re-test your model choice periodically. A model that lagged behind six months ago may now be competitive or better for your use case; see our main AI models guide as a starting point, but verify against current versions.
FAQ
Why are AI labs releasing so many models instead of one big annual update? Competitive pressure and faster training and evaluation cycles have pushed most major labs toward frequent, incremental releases rather than a single yearly flagship, similar to how mobile apps ship frequent updates rather than annual versions.
Which lab has released the most models in 2026? Multiple labs, both proprietary and open-weight, have shipped several model generations this year rather than a single release, and the pace has been broadly similar across OpenAI, Anthropic, Google, and the leading open-weight labs.
Is a newer model version always better? Usually better on benchmarks, but not always better for your specific task. It's worth testing a new release against your actual use case rather than assuming a higher version number means a better fit.
How often is this timeline updated? This page reflects major releases through late July 2026 and will be updated as new significant models ship. For the most current lineup at any given moment, check each provider's official site directly.
Last updated: July 30, 2026. This timeline covers major releases through late July 2026 and will be revised as new models ship; always confirm the current lineup on each provider's official site.
