feat(stackit): add GPT-OSS 20B and Qwen3.6 27B models

[amankalra172]

Summary

• Add openai/gpt-oss-20b: 131K context, 8192 max output, tool calling + reasoning, $0.16/$0.27 per 1M tokens
• Add Qwen/Qwen3.6-27B: 262K context, 16384 max output, tool calling (reasoning disabled on this STACKIT deployment), $0.49/$0.71 per 1M tokens

Both models are now available on STACKIT AI Model Serving. Prices are converted from EUR (0.15€/0.25€ and 0.45€/0.65€ respectively) using the same conversion rate applied to existing STACKIT models.

Test plan

• Verify model IDs match STACKIT API (openai/gpt-oss-20b, Qwen/Qwen3.6-27B)
• Verify pricing against STACKIT pricing page
• Confirm context and output limits from STACKIT docs

[rekram1-node]

lets use base_model syntax if possible for both of these

[fheuer]

possible duplicate of: #2029

[amankalra172]

Thanks for the suggestion! We can't use base_model here because STACKIT's deployments have provider-specific constraints that differ from the canonical models:

• Qwen/Qwen3.6-27B: the base alibaba/qwen3.6-27b has reasoning=true, multimodal input (image/video/audio), output=65_536, and costs $0.6/$3.6/M. STACKIT's deployment has reasoning disabled, text-only input, output=16_384, and different pricing.
• openai/gpt-oss-20b: no canonical source exists in models.dev yet.

This is also consistent with how all existing STACKIT models are defined — gpt-oss-120b, Qwen3-VL-235B, Llama-3.3-70B, etc. all use full specs for the same reason. Using base_model would inherit incorrect specs for this provider.

[rekram1-node]

u can override any details, base_model is used for linking and referencing in the website.