feat(groq): add Qwen3.8 27B

[DevSuyashA]

Adds the Qwen3.8 27B model to the Groq provider catalog.

Model data (reported by GET /v1/models):

• base_model: alibaba/qwen3.8-27b
• context window: 131,042
• max output tokens: 16,384
• cost (USD/MTok): input 0.80, output 4.00, cache_read 0.40
• modalities (input): text, image
• reasoning_options: effort [none, default]

Validated with bun validate (import gate passes).

Sources (first-party, accessed 2026-08-29):

• https://console.groq.com/docs/models — context/output limits, modalities, model capabilities
• https://console.groq.com/docs/reasoning — reasoning_effort options
• https://groq.com/pricing — input/output/cache_read pricing

Context limit note: context_limit 131,042 confirmed live against GET https://api.groq.com/openai/v1/models for qwen/qwen3.8-27b on 2026-08-29 (returned context_window = 131042). The value is now closed/confirmed — no correction needed.

[github-actions]

Action items

• [medium] [possible mistake] providers/groq/models/qwen/qwen3.8-27b.toml:10 - Check: Provider context limit must match what Groq actually serves. Why: context = 131_042 is unique in the catalog and differs from every other Groq chat model (including the same-surface peer qwen/qwen3.6-27b), which use 131_072. That is a plausible typo or misread of a power-of-two window. Action: Confirm the value against Groq’s model page or live /v1/models for qwen/qwen3.8-27b; if the host is 131,072 tokens, correct it to 131_072.
• [low] [possible mistake] providers/groq/models/qwen/qwen3.8-27b.toml:4 - Check: Data-changing PRs should cite first-party pricing/docs/API for material facts. Why: Cost (0.80 / 4.00 / cache_read 0.40), limits, and modalities cannot be reviewed from the catalog alone; the PR body only asserts API verification without a mapped Groq pricing or model doc URL. Action: Add a leading top-of-file comment (and/or PR body) with direct Groq sources and what each supports (pricing, context/output, modalities, reasoning_effort values).

[github-actions]

Action items

• [medium] [possible mistake] providers/groq/models/qwen/qwen3.8-27b.toml:10 - Check: Provider limit.context should match this host’s published/API context window. Why: context = 131_042 is not a power-of-two window and differs from the closest Groq peer (qwen/qwen3.6-27b uses 131_072) and every other standard Groq text model at 128k (131_072). That pattern strongly suggests a typo (e.g. 0 vs 7) rather than an intentional host limit. Action: Confirm against Groq’s model docs or /v1/models and set context to the real value (likely 131_072 unless Groq truly advertises 131_042).