Skip to content

Feature/zoo 608 add glm 5 2 support#110

Merged
krzychdre merged 3 commits into
mainfrom
feature/zoo-608-add-glm-5-2-support
Jun 17, 2026
Merged

Feature/zoo 608 add glm 5 2 support#110
krzychdre merged 3 commits into
mainfrom
feature/zoo-608-add-glm-5-2-support

Conversation

@krzychdre

Copy link
Copy Markdown
Owner

No description provided.

krzychdre and others added 3 commits June 17, 2026 23:51
…#555)

Port of Zoo-Code PR #555. Registers `claude-fable-5` across the Anthropic,
Bedrock, Vertex, OpenRouter, Requesty and Vercel AI Gateway providers
(context 1M, direct maxTokens 128k / 8k on Bedrock+Vertex, adaptive-binary
reasoning, temperature unsupported), and routes the model through adaptive
thinking.

Fork-specific: our Vertex handler never received the adaptive-binary reasoning
routing that Zoo already had, so it emitted `{type:"enabled",budget_tokens}` for
Opus 4.7/4.8 and Fable 5. Recompute `thinking` via getAnthropicProviderReasoning
in AnthropicVertexHandler.getModel so Vertex matches the direct provider.

Co-authored-by: T <taltas@users.noreply.github.com>
Co-authored-by: Elliott de Launay <edelauna@gmail.com>
…ng streams (zoo #588)

Replace the duplicated inline reasoning_content extraction block in the
OpenAI-compatible streaming providers (deepseek, openai, qwen-code,
requesty, unbound) with the shared extractReasoningFromDelta helper, which
prefers delta.reasoning_content and falls back to delta.reasoning
(OpenRouter style). Adds streaming coverage for both paths.

Fork adaptations: the helper already existed here; mimo and opencode-go do
not exist in this fork and are skipped; unbound has no spec file here so its
new streaming tests are skipped (the helper has its own unit tests).

Co-authored-by: dw <41457565+daewoongoh@users.noreply.github.com>
Co-authored-by: Oh Daewoong <dw.oh@samsung.com>
Adds the glm-5.2 flagship model (1M context) to both Z.ai model maps with a
new Max reasoning-effort tier; High is the default and Max is opt-in. Persisted
reasoning effort not offered by the current model now falls back to the model
default instead of silently disabling reasoning, and the streaming create() is
wrapped in handleOpenAIError for parity with the base class. GLM-5.2 pricing is
provisional (mirrors GLM-5.1). Adds Max translations across all 18 locales.

Co-authored-by: Mob Code 100 <66469454+MobCode100@users.noreply.github.com>
@krzychdre krzychdre merged commit fda381c into main Jun 17, 2026
4 of 6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant