Inkling 2 Small and Gemma 4

Three new open-weight models are in the model picker starting today, on every plan: the second generation of Inkling, Inkling 2 Small from Thinking Machines Lab, and two Gemma 4 models from Google.
What you can do with them
- Code and build. Inkling 2 Small is tuned for coding, tool use and knowledge work, so it's a good fit for debugging, refactoring or planning a project step by step.
- Work with big inputs. Inkling 2 Small has a one million token context window, so a long codebase, a contract or a whole chat history fits in one conversation.
- Ask about a picture. All three read images. Attach a screenshot, a diagram or a photo of a page and ask what it shows.
- Choose how hard it thinks. On Inkling 2 Small, pick no reasoning for quick replies, or low, medium or high when a problem needs more care.
- Look things up. Web search runs through Perplexity, under Perplexity's own terms and privacy policy.
Second-gen Inkling, at a lower cost
Inkling 2 Small sends start at 2 base requests, half of the original Inkling's 4. It also has four times the context window: one million tokens instead of 256,000. Longer chats and higher reasoning settings use more.
The original Inkling stays in the picker, so you can keep using it.
Gemma 4 from Google
Gemma 4 is Google's open-weight model family. Both Gemma 4 models start at 1 base request per send, read images and have a 256,000 token context window. They answer directly, with no reasoning setting to choose.
- Gemma 4 31B is the larger of the two, for everyday chat, writing and questions about images.
- Gemma 4 26B A4B is a mixture-of-experts model. Only about 4 billion of its parameters are active for each reply, so it's fast.
Where they run, and what happens to your chats
All three are open-weight models. Their makers have published the models themselves, so independent companies can run them on their own servers.
On just4o.chat:
- Inkling 2 Small runs on Together AI and Modal in the United States, and on Nebius in the European Union.
- Gemma 4 31B runs on Parasail.
- Gemma 4 26B A4B runs on Google Vertex AI, with Parasail as a backup.
Requests are sent through the Vercel AI Gateway with zero data retention, and your chats are not used for training.
Your conversation history still lives in your just4o.chat account, the same as with every other model, so you can come back to it later.
At a glance
| Inkling 2 Small | Gemma 4 31B | Gemma 4 26B A4B | |
|---|---|---|---|
| Cost per send | From 2 base requests | From 1 base request | From 1 base request |
| Image input | Yes | Yes | Yes |
| Reasoning | None, low, medium or high | No setting | No setting |
| Web search | Yes, through Perplexity | Yes, through Perplexity | Yes, through Perplexity |
| Context window | One million tokens | 256,000 tokens | 256,000 tokens |
| Privacy level | Zero data retention | Zero data retention | Zero data retention |
| Plans | Every plan | Every plan | Every plan |
Try it
Open chat, choose Inkling 2 Small or a Gemma 4 model in the model picker, and hand it a coding problem, a long document or a screenshot. The full details are on the Inkling 2 Small, Gemma 4 31B and Gemma 4 26B A4B model pages.

