x.com/_xjdr/status/2090392818531381387
1 correction found
No one is releasing base models anymore.
This absolute claim is false: companies were still releasing base/pretrained models recently, including IBM in April 2026 and Ai2 with Olmo 3 base checkpoints.
Full reasoning
The claim uses an absolute (“No one”), so a single clear counterexample is enough to show it is incorrect.
Recent official releases contradict it:
- IBM Research announced on April 29, 2026 that Granite 4.1 language models were released in “3B, 8B, and 30B parameter base and instruct model sizes.” That is a direct, recent release of base models.
- Ai2's official Olmo 3 release says “Olmo 3-Base (7B, 32B) is our most powerful base model yet” and also states “Instead of releasing only the final weights, we provide checkpoints from each major training milestone: the base pretrained model...” That is another direct release of base/pretrained models.
Because multiple organizations officially released base models/checkpoints in the recent period, the blanket statement that no one is doing so anymore is factually wrong.
3 sources
- Introducing the IBM Granite 4.1 family of models - IBM Research
At the heart of Granite 4.1 is a new generation of dense, decoder-only language models, offered in 3B, 8B, and 30B parameter base and instruct model sizes.
- Olmo 3: Charting a path through the model flow to lead open-source AI | Ai2
Olmo 3-Base (7B, 32B) is our most powerful base model yet... Instead of releasing only the final weights, we provide checkpoints from each major training milestone: the base pretrained model...
- Introducing Gemma 3 270M: The compact model for hyper-efficient AI - Google Developers Blog
An instruction-tuned model is released alongside a pre-trained checkpoint... We are releasing both pretrained and instruction tuned models.