feat(alibaba-token-plan-cn): add DeepSeek V4.1 Flash

[netphils]

Summary

Add deepseek-v4.1-flash — DeepSeek's September 2026 flash release, now on the Alibaba Token Plan (China) model list — to providers/alibaba-token-plan-cn. The entry is override-only and references the existing lab metadata via base_model.

Changes

• providers/alibaba-token-plan-cn/models/deepseek-v4.1-flash.toml (new)
  • base_model = "deepseek/deepseek-v4.1-flash"
  • reasoning_options: toggle + effort ["low", "high", "max"]
  • interleaved.field = "reasoning_content"
  • cost: 0 input / 0 output / 0 cache_read / 0 cache_write (Credits-based subscription)
• No models/deepseek/ change required — the lab entry models/deepseek/deepseek-v4.1-flash.toml already exists (1M context / 384K output, MIT, open weights).

Evidence

• Model availability on Token Plan Personal Edition CN (lists deepseek-v4.1-flash, "推理模型、视觉理解、文本生成"): https://help.aliyun.com/zh/model-studio/token-plan-personal-overview
• Reasoning wire format on this host — enable_thinking toggle; reasoning_effort accepts low / high (default) / max, with medium and xhigh mapping to high; reasoning streams in reasoning_content: https://platform.qianwenai.com/docs/api-reference/chat/openai-chat (accessed 2026-09-14)
• Beijing-region endpoint + Responses API support for deepseek-v4.1-flash (this plan is Beijing-only): https://www.alibabacloud.com/help/en/model-studio/deepseek-api

Validation

• bun validate passes locally (exit 0) after the change.

Review notes

• Unlike the sibling deepseek-v4-flash.toml (high/max only), low is included here because the current host docs list low as a distinct effort level specifically for deepseek-v4.1-flash, deepseek-v4-flash-0731, and deepseek-v4-pro-0813.
• thinking_budget is intentionally omitted: docs mark it "not settable" for this model.

[github-actions]

No actionable findings.