And another one: three top models in four weeks
GPT-5.2 appears on 11 December 2025, Gemini 3 Flash six days later. Together with November, that makes three large models in four weeks — from three different houses.
December 2025 continued what November had begun. On 11 December, OpenAI released GPT-5.2; on 17 December, Google presented Gemini 3 Flash — a smaller, faster model alongside the Gemini 3 Pro released in November.
The release date of GPT-5.2 is recorded in the title of the accompanying system card: OpenAI, GPT-5.2 System Card (11. Dezember 2025). Documents of that kind are more reliable evidence than news reports — they come from the manufacturer and carry a date.
Four weeks, three houses
- 18 November 2025: Gemini 3 Pro (Google)
- 24 November 2025: Claude Opus 4.5 (Anthropic)
- 11 December 2025: GPT-5.2 (OpenAI)
- 17 December 2025: Gemini 3 Flash (Google)
Anyone who had set up a comparative test of the large models in the autumn of 2025 was free to repeat it in December. By the third time, you start wondering what for.
The side effect that is talked about less
Alongside the large models, smaller ones appeared during this period as well: Gemini 3 Flash is exactly that. For businesses, that is often the more important news — not every task needs the strongest model.
Reading a piece of evidence and placing it in context is not a doctoral thesis. Using the most expensive model for it is like fetching bread with a lorry.
In our own systems, two models therefore run side by side: the strong one for documents, a smaller and cheaper one for emails. The split saves money without quality suffering anywhere — because the tasks differ in difficulty.
Sources
This is how our system does it