From 4796eb87bd152bd084651f0c462a0696a3a808ba Mon Sep 17 00:00:00 2001 From: pierreraby Date: Fri, 13 Nov 2026 13:21:03 +0100 Subject: [PATCH] chore(release): prepare 0.8.3 Patch release carrying two user-facing fixes and one pricing refresh: - Responses wire repair for zai-org/GLM-5.3 and mistral/mistral-large-4 (#215, reported in #214). - Reviewed zero pricing for thinkingmachines/inkling-2-small:free (#213). --- CHANGELOG.md | 13 +++++++++---- package-lock.json | 4 ++-- package.json | 2 +- 3 files changed, 12 insertions(+), 7 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index f05091d..055262b 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,11 +1,16 @@ # Changelog -## Unreleased +## 0.8.3 - 2026-11-13 -- Tag the Responses `system`/`developer` and `user` items with the item-level `type: "message"` the Provider API adapter requires for `zai-org/GLM-5.3`, which answered `400 invalid_request: input item is missing type` on the first turn. The rewrite is limited to the models known to need the verbose form, so every other model keeps the exact body the host sends. ([#214](https://github.com/patlux/pi-commandcode-provider/issues/214)) -- Drop the inert `reasoning` block pi sends when thinking is off for `zai-org/GLM-5.3` and `mistral/mistral-large-4`, which reject `effort: "none"` with `unsupported output_config.effort` and `reasoning_effort none is not supported for this model`. Omitting the field succeeds; declaring `off: null` would clamp to the lowest allowed level and silently force reasoning on. The repair table is per-model and evidence-driven, so no other model's request body changes. ([#214](https://github.com/patlux/pi-commandcode-provider/issues/214)) +- Tag the Responses `system`/`developer` and `user` items with the item-level `type: "message"` the Provider API adapter requires for `zai-org/GLM-5.3`, which answered `400 invalid_request: input item is missing type` on the first turn. The rewrite is limited to the models known to need the verbose form, so every other model keeps the exact body the host sends. ([#214](https://github.com/patlux/pi-commandcode-provider/issues/214), [#215](https://github.com/patlux/pi-commandcode-provider/pull/215)) +- Drop the inert `reasoning` block pi sends when thinking is off for `zai-org/GLM-5.3` and `mistral/mistral-large-4`, which reject `effort: "none"` with `unsupported output_config.effort` and `reasoning_effort none is not supported for this model`. Omitting the field succeeds; declaring `off: null` would clamp to the lowest allowed level and silently force reasoning on. The repair table is per-model and evidence-driven, so no other model's request body changes. ([#214](https://github.com/patlux/pi-commandcode-provider/issues/214), [#215](https://github.com/patlux/pi-commandcode-provider/pull/215)) +- Add the live Provider API addition `thinkingmachines/inkling-2-small:free` to the reviewed pricing snapshot at zero cost, free while the free tier lasts with no fixed expiry published, and refresh the 98-model API and pricing snapshots to 2026-11-13. The model is already usable through the live `/provider/v1/models` response with a zero estimate and an unknown-price warning; static capability metadata stays at `command-code@1.88.5`, so it remains text-only without reasoning or effort flags until upstream ships it. ([#213](https://github.com/patlux/pi-commandcode-provider/pull/213)) -- Add the live Provider API addition `thinkingmachines/inkling-2-small:free` to the reviewed pricing snapshot at zero cost, free while the free tier lasts with no fixed expiry published, and refresh the 98-model API and pricing snapshots to 2026-11-13. The model is already usable through the live `/provider/v1/models` response with a zero estimate and an unknown-price warning; static capability metadata stays at `command-code@1.88.5`, so it remains text-only without reasoning or effort flags until upstream ships it. +### Contributors + +- @ThomasByr — reported the `zai-org/GLM-5.3` failures on the Responses endpoint with a minimal reproduction ([#214](https://github.com/patlux/pi-commandcode-provider/issues/214)). +- @sandexzx — refreshed the reviewed pricing snapshot for `thinkingmachines/inkling-2-small:free` ([#213](https://github.com/patlux/pi-commandcode-provider/pull/213)). +- @pierreraby — isolated and proved both wire causes against the live Provider API, swept all 74 `/responses` models to bound the blast radius, implemented the per-model repair table with tests, and prepared this release. ## 0.8.2 - 2026-11-11 diff --git a/package-lock.json b/package-lock.json index 30ee773..ddf1a68 100644 --- a/package-lock.json +++ b/package-lock.json @@ -1,12 +1,12 @@ { "name": "pi-commandcode-provider", - "version": "0.8.2", + "version": "0.8.3", "lockfileVersion": 3, "requires": true, "packages": { "": { "name": "pi-commandcode-provider", - "version": "0.8.2", + "version": "0.8.3", "license": "MIT", "devDependencies": { "@types/node": "25.6.0", diff --git a/package.json b/package.json index ec8804d..6b26850 100644 --- a/package.json +++ b/package.json @@ -1,6 +1,6 @@ { "name": "pi-commandcode-provider", - "version": "0.8.2", + "version": "0.8.3", "description": "pi custom provider for Command Code API (commandcode.ai)", "type": "module", "keywords": [