Skip to content

feat(zenmux): add custom/deepseek/deepseek-v4-flash - #3898

Open
xiaojiezj wants to merge 1 commit into
anomalyco:devfrom
xiaojiezj:add-zenmux-custom-deepseek-v4-flash
Open

feat(zenmux): add custom/deepseek/deepseek-v4-flash#3898
xiaojiezj wants to merge 1 commit into
anomalyco:devfrom
xiaojiezj:add-zenmux-custom-deepseek-v4-flash

Conversation

@xiaojiezj

Copy link
Copy Markdown

Add custom/deepseek/deepseek-v4-flash to ZenMux provider.

  • base_model = "deepseek/deepseek-v4-flash"
  • Reasoning support with toggle + effort options
  • Cost: input $0.14, output $0.28, cache_read $0.0028 per million tokens

Data sourced from https://zenmux.ai/api/anthropic/v1/models

@github-actions

Copy link
Copy Markdown
Contributor

Action items

  • [high] [possible mistake] providers/zenmux/models/custom/deepseek/deepseek-v4-flash.toml:2 - Check: reasoning_options must match the provider’s Anthropic request surface, not an OpenAI-compatible sibling. Why: This entry sets npm = "@ai-sdk/anthropic" and api = "https://zenmux.ai/api/anthropic/v1", but copies toggle + effort low|medium|high from the OpenAI-compatible deepseek/deepseek-v4-flash entry. ZenMux’s own Anthropic notes document thinking.type enabled|disabled with required budget_tokens (≥1024), not effort; native DeepSeek Anthropic documents output_config.effort = high|max (budget ignored). The proposed effort values match neither surface. Action: Re-audit against ZenMux Anthropic reasoning docs / a real Anthropic Messages request for this model ID, then set options to the verified controls (likely toggle + budget_tokens, or native DeepSeek’s toggle + effort high|max if ZenMux passthrough is confirmed). Do not copy the OpenAI-compatible sibling.
  • [medium] [possible mistake] providers/zenmux/models/custom/deepseek/deepseek-v4-flash.toml:4 - Check: interleaved must describe this inference surface’s response shape. Why: field = "reasoning_content" is an OpenAI-style field and matches the OpenAI-compatible ZenMux DeepSeek entry, but no ZenMux Anthropic-surface models (@ai-sdk/anthropic + /api/anthropic/v1) declare interleaved; Anthropic Messages exposes thinking blocks, not reasoning_content. Action: Confirm the Anthropic endpoint actually returns interleaved reasoning_content; if not, remove [interleaved].

@rekram1-node

Copy link
Copy Markdown
Collaborator

/oc fix this pr, interleaved is incorrect here, if it is using the anthropic api then reasoning_content isnt valid.

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.

2 participants