tashfeenahmed/freellmapi v0.16.0

on GitHub

FreeLLMAPI v0.16.0 adds decision models and two new routing controls, removes the retired GitHub Models provider, and fixes LLMTR keys that older versions switched off.

Decision models

• New: decision models. They answer typed questions about a piece of text (pick an option, yes/no, or a score) with probabilities instead of generated text. They come from the catalog and use your existing provider keys, so there's nothing new to add.
  • API: POST /v1/decisions (also /v1/systemone) and GET /v1/decisions/models.
  • Dashboard: a new Models → Decision tab with on/off toggles and a "Try it" box.
  • Chat routing never uses them. (#1537)

Routing

• Minimum context size: one setting (off, 32k, 128k, 512k or 1m) keeps models with smaller context windows out of routing. It's off by default. GET/PUT /api/settings/min-context-window. (#1445)
• Claude Code tiers: map opus, sonnet, haiku or default to tier:pro, tier:mid or tier:normal. That routes across every model in that tier instead of one pinned model. (#1444)
• Custom endpoints: a local server such as LM Studio that rejects a tool_choice naming one function now gets one retry with "required" instead of failing. Thanks @rubicon. (#1549)
• A maximum output length learned from one custom endpoint no longer limits other endpoints that serve the same model. (#1532)

Providers

• GitHub Models is removed. GitHub retired it on 30 July 2026, and its server now answers every request with "OK", so keys looked healthy while every request failed. Leftover GitHub keys and custom GitHub models are removed when you update. (#1560)
• Key checks: a provider that answers 200 with something other than JSON now shows an error instead of "healthy". This won't switch the key off. (#1560)
• LLMTR:
  • Keys that older versions switched off by mistake are switched back on.
  • Pasting in a new key turns it on.
  • "Unchecked" is now shown in grey with an explanation instead of as an error. (#1551)
• Groq: reasoning_effort values Groq doesn't accept (none, minimal) become low, so the request stays on Groq. Thanks @silvertakana. (#1480)
• Cloudflare: images in vision requests are now passed through instead of being dropped. Thanks @The88788. (#1462)

Dashboard

• The Keys and Models pages no longer get wider than the screen on phones. (#1537, #1551)

Updating

Desktop users on v0.13.5 or later can update from inside the app: ⋯ → Check for updates. Docker: ghcr.io/tashfeenahmed/freellmapi:v0.16.0. The latest tag also points to this release.

Existing keys and profiles are retained, except GitHub Models keys, which are removed because the service no longer exists. This release has three database migrations, which run automatically: a log table for decision requests, switching affected LLMTR keys back on, and removing GitHub Models rows. All three can be undone.

Full changelog

⭐ Like the free router? Go Premium, the live signed catalog, $19/yr, cancel anytime.