fix: [missing-model] Groq: qwen/qwen3.8-27b

[github-actions]

Summary

Adds the missing Groq-hosted model qwen/qwen3.8-27b, which Groq lists as a preview model but was absent from this catalog. Lab metadata already exists at models/alibaba/qwen3.8-27b.toml, so the provider entry is override-only via base_model.

Changes

• Added providers/groq/models/qwen/qwen3.8-27b.toml
  • base_model = alibaba/qwen3.8-27b
  • cost.input = 0.80, cost.output = 4.00 (USD per million tokens)
  • limit.context = 131_042, limit.output = 16_384
  • modalities.input = ["text", "image"], modalities.output = ["text"] (Groq host surface; lab also supports video)
  • reasoning_options = effort none | default | low | medium | high

Evidence

Claim	Source	What it establishes
Model ID qwen/qwen3.8-27b is live on Groq (preview)	Groq models list, model page	Listed under Preview Models with ID qwen/qwen3.8-27b
Pricing $0.80 input / $4.00 output per 1M tokens	model page pricing	Input $0.80, output $4.00
Context 131,042; max output 16,384	model page limits, models table	CONTEXT WINDOW 131,042; MAX OUTPUT TOKENS 16,384
Input text + images; output text	model page	INPUT: Text, images; OUTPUT: Text; Vision capability
Reasoning effort none, default, low, medium, high	Groq reasoning docs, model best practices	Dedicated Qwen 3.8 27B effort table; best practices mention none/default and low/medium/high
No prompt-cache price on this model	Prompt caching	Supported models are GPT-OSS only; Qwen not listed
Lab base already present	models/alibaba/qwen3.8-27b.toml	No new lab file required

Validation

• Cross-checked Groq first-party model page, models index, reasoning docs, and prompt-caching docs against the peer Groq entry qwen/qwen3.6-27b.
• Did not run bun validate (shell disabled in this workflow).

Review notes

• Did not set status = "beta" even though Groq labels the model Preview; the existing peer qwen3.6-27b is also Preview and has no status field. Only a few other Groq preview models use status = "beta".
• Did not author cache_read; Groq prompt caching currently applies only to GPT-OSS models (the existing qwen3.6-27b cache_read = 0.30 looks inconsistent with current docs and was left untouched).
• Context is intentionally 131_042 (Groq’s documented value), not the lab’s 262_144.

Closes #5724

Automated by the issue fixer: https://github.com/anomalyco/models.dev/actions/runs/33170791128