feat: add gpt-oss family + Thinking Machines Inkling

[brunobuddy]

💭 Why

The catalog had no OpenAI open-weight models and no Thinking Machines provider. The two Gemma 4 entries were also missing the thinking toggle Google documents for them.

✨ What changed

• Add openai/gpt-oss-120b, gpt-oss-20b, gpt-oss-safeguard-120b, gpt-oss-safeguard-20b (API-key).
• Add thinking-machines provider with Inkling, from Tinker's OpenAI-compatible endpoint docs.
• Add generationConfig.thinkingConfig.thinkingLevel (minimal/high) to both gemma-4 entries.
• Commit codegen output for the modelparams package.

Type of change

• Add a model (new YAML file under models/)
• Add a provider (new folder under models/)
• Add or update parameters on an existing model

Source

• gpt-oss: gpt-oss repo (temperature/top_p 1.0 recommendation), harmony docs (reasoning effort low/medium/high, default medium), model page (131,072 max output; function calling and structured outputs supported), safeguard guide.
• Inkling: Tinker OpenAI-compatible API (max_tokens, temperature, top_p, reasoning_effort presets with 0.9 default, separate_reasoning), thinking effort, model card.
• Gemma 4: Gemma on the Gemini API (thinking_level "high" enables, "minimal" disables; no default documented, so none listed).

📝 Notes

• Gemma 3 / 3n were left out on purpose: Google silently removed them from the Gemini API around May 2026 (ListModels no longer returns them), so there is no current API contract to catalog. gemma-4-12b-it and E2B/E4B are open weights only, not servable on the API.
• No logo for thinking-machines yet; the site falls back to the generic mark. lobehub icons has no mark for them either.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
modelparams.dev	Ready	Preview	Jul 20, 2026 9:31am