Skip to content

Add deepseek-v4.1-flash to alibaba-cn provider - #7396

Merged
rekram1-node merged 1 commit into
anomalyco:devfrom
Spin-Particle:feat/alibaba-cn-deepseek-v4-1-flash
Sep 18, 2026
Merged

rekram1-node merged 1 commit into
anomalyco:devfrom
Spin-Particle:feat/alibaba-cn-deepseek-v4-1-flash

Conversation

@Spin-Particle

Copy link
Copy Markdown
Contributor

Adds DeepSeek V4.1 Flash to the Alibaba Bailian pay-as-you-go provider (alibaba-cn, Beijing), using the existing lab entry models/deepseek/deepseek-v4.1-flash.toml via base_model.

Sources (accessed 2026-09-18):

Host-specific deltas vs the lab entry:

  • Hybrid reasoning with both controls. enable_thinking = true|false toggles thinking (the deepseek doc classifies the v4 family as hybrid-thinking models on this host), and reasoning_effort accepts low / high / max with default high — the doc's capability table explicitly lists low as supported for deepseek-v4.1-flash (only v4.1-flash, v4-flash-0731 and v4-pro-0813 support low). reasoning_content streams in deltas.
  • Cost: peak/off-peak (峰谷) pricing. Beijing lists CNY 2/1 (peak/off-peak) input and 8/4 output per 1M tokens with context-cache discount. Following the existing convention on this host (deepseek-v4-flash-0731 / deepseek-v4-pro-0813 record the peak price), this entry records the peak price converted to USD: 0.278 / 1.111 / 0.014 (cache read).
  • Responses API supported on Beijing and Singapore only; chat completions available everywhere.
  • Limits and modalities are identical to the lab entry (1M context, 384_000 output, text+image input), so they are inherited and not restated (override-only). Note the host's own default max_tokens is 393_216 shared with thinking, but the model's max output is 384_000 per the lab entry.

Not added to alibaba (international) — the international gateway's model list prices in a different rate card (Singapore peak CNY 2.188/8.75); it can be added separately once pricing is confirmed. Also not added to token-plan / coding-plan catalogs — the Beijing pay-as-you-go catalog is what the model market page lists.

Related: #7148 added v4.1-flash to alibaba-token-plan (Singapore) earlier; this fills the Beijing pay-as-you-go gap.

bun validate passes.

@github-actions

Copy link
Copy Markdown
Contributor

Action items

  • [medium] [violation] providers/alibaba-cn/models/deepseek-v4.1-flash.toml:17 - Check: CNY costs must be converted to USD/MTok with rate and date in a leading top-of-file comment. Why: The file authors USD 0.278 / 1.111 / 0.014 from Beijing CNY peak prices but never records the FX rate, date, or rate source, so the conversion cannot be audited or reproduced. Action: Add a leading comment with the CNY→USD rate, date, and source (same pattern as qwen3.8-flash.toml / qwen3.7-flash.toml on this host), e.g. # CNY→USD rate: X.XXXX, YYYY-MM-DD, source: ….

@github-actions github-actions Bot added the reviewer: ready Automated review found no actionable items label Sep 18, 2026
@rekram1-node
rekram1-node merged commit 4056f86 into anomalyco:dev Sep 18, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

reviewer: ready Automated review found no actionable items

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants