OpenAI
GPT
NOT YET
GPT-6-mini, expected in 17 days
last out GPT-6 on 12 Sep 2026, 6 days ago
history · OpenAI
8 releases · 96.4 d avg · 288 d max
Anthropic
Claude
NOT YET
Claude Fable 5.2, expected in 14 days
last out Claude Fable 5.1 on 1 Sep 2026, 17 days ago
history · Anthropic
2 releases · 75.0 d avg · 75 d max
xAI
Grok
NOT YET
Grok 4.7, expected in 27 days
last out Grok 4.6 on 12 Aug 2026, 37 days ago
history · xAI
3 releases · 52.0 d avg · 69 d max
DeepSeek
DeepSeek
OUT
nothing announced yet
last out DeepSeek Pro on 14 Sep 2026, 4 days ago
history · DeepSeek
3 releases · 38.0 d avg · 76 d max
Alibaba
Qwen
NOT YET
Qwen 4 Max, expected in 20 days
last out Qwen 3.9 Max on 25 Aug 2026, 24 days ago
history · Alibaba
2 releases · 98.0 d avg · 98 d max
Z.ai
GLM
OUT
nothing announced yet
last out GLM 5.3 Flash on 26 Aug 2026, 23 days ago
history · Z.ai
3 releases · 35.5 d avg · 63 d max
the timeline
The light version of GPT-6, built to answer fast where the full model thinks for a while. Nothing spectacular on paper, but this is the kind of model that ends up inside everyday products.
benchmark leak
in 17 days
The first real step up for the family: context goes to one million tokens, enough to swallow a whole codebase or a full report without splitting it.
developer conference
in 10 days
The release that makes GPT-6 public: steadier on long reasoning and finally comfortable with tools. This is the one replacing GPT-5.6 in everyday use.
official OpenAI blog
The open preview before the official release: same foundations, a bit more talkative, with a clear jump in accuracy. This is what early testers got their hands on.
beta programme
The cheap one, designed for high-volume calls. It answers faster and costs far less, with a small loss as soon as a task needs reasoning.
OpenAI API
An internal build that never shipped, where the lab tested a new attention architecture. It serves as an internal reference, not a product.
internal note
The last GPT-5.6 upgrade before GPT-6: context reaches one million tokens and reasoning gains a little more. This is the version that held the fort all summer.
official OpenAI blog
The first version genuinely at ease with tools: it chains several calls without losing the thread, which opened up agent use cases.
official OpenAI blog
A transition generation: longer memory and steadier reasoning, without changing anyone's habits.
technical note
Mostly a safety update: guardrails were tightened after a series of public jailbreaks.
technical note
The model that took this technology mainstream: for the first time you could show it a picture and ask it to talk about it. No comparable numbers, this is where the line starts.
archive
Count page views, show a public audience. No ads today, no cross-site tracking. Details in privacy.