fix: Add qwen3.8-max (GA) to alibaba-token-plan / alibaba-token-plan-cn providers

[github-actions]

Summary

Alibaba Token Plan Personal docs now list GA qwen3.8-max (marked NEW) next to qwen3.8-max-preview. The catalog only had the preview on alibaba-token-plan / alibaba-token-plan-cn, so Token Plan users could not pick the GA model. This adds the lab metadata and override-only provider entries for both Token Plan hosts.

Changes

• Added models/alibaba/qwen3.8-max.toml
  • name: Qwen3.8 Max
  • multimodal flagship (text/image/video → text)
  • context = 1_000_000, output = 131_072
  • release_date / last_updated = 2026-08-03
  • attachment, reasoning, tool_call = true; open_weights = false
• Added providers/alibaba-token-plan/models/qwen3.8-max.toml
  • base_model = "alibaba/qwen3.8-max"
  • no status = "beta" (GA)
  • structured_output = true
  • reasoning_options: toggle + effort low/medium/xhigh + budget_tokens 0..262144
  • interleaved.field = "reasoning_content"
  • Credits plan cost zeros
• Added providers/alibaba-token-plan-cn/models/qwen3.8-max.toml (same shape as intl)

Unlike always-on preview, GA is hybrid thinking (enable_thinking on/off), so provider entries include toggle plus the preview-style effort/budget controls.

Evidence

Claim	Source
Token Plan lists qwen3.8-max (GA, NEW) and preview	CN Token Plan Personal, EN Token Plan Personal, QwenCloud Token Plan Personal
Context 1M, max output 131K, max reasoning 262K; text/image/video in, text out; structured outputs; function calling	QwenCloud model page, 千问 model page
GA hybrid thinking; effort low/medium/xhigh (default xhigh); budget mutually exclusive with effort	OpenCode (intl), OpenCode (CN), Thinking docs
Preview remains always-on (no toggle)	Same OpenCode guides (contrast section for qwen3.8-max-preview)
Model ID on Model Studio catalog	help.aliyun.com models

Validation

• Cross-checked Token Plan allowlists (CN + EN + QwenCloud), model pages, thinking docs, and OpenCode client guides.
• Mirrored existing preview/Token Plan TOML patterns (base_model, Credits zero cost, interleaved, structured_output override).
• Did not run bun validate (no shell per instructions).

Review notes

• Did not add qwen3.8-max to pay-as-you-go alibaba / alibaba-cn (or other hosts); issue scope is Token Plan only, even though Model Studio now lists the ID more broadly.
• open_weights = false: launch copy discusses open-sourcing Max-class weights, but no verified public weight drop was used here; matches existing opencode-go entry.
• Token Plan cost remains 0 (Credits subscription), same as sibling models.
• Provider npm is OpenAI-compatible; OpenCode examples also use Anthropic Messages with thinking.type — catalog reasoning_options follow the Chat/enable_thinking surface documented for this host, consistent with other hybrid Qwen Token Plan entries.

Closes #3981

Automated by the issue fixer: https://github.com/anomalyco/models.dev/actions/runs/30801434421

[MstKenway]

Please also add a new config file for deepseek-v4-flash-0731, which is different from deepseek-v4-flash. Thanks!