[speteai]

Adds AI Power Grid (aipg), a network of independent GPU operators behind one
OpenAI-compatible API at https://api.aipowergrid.io/v1. I am a community developer building on the grid, not
part of its team.

Files

• providers/aipg/provider.toml: @ai-sdk/openai-compatible, AIPG_API_KEY (the name the grid's own
  integration docs use).
• providers/aipg/logo.svg: single-colour version of the project's cube logo (currentColor, square viewBox).
• providers/aipg/models/gpt-oss-120b.toml → openai/gpt-oss-120b
• providers/aipg/models/qwen38-flash-next-nvfp4.toml → alibaba/qwen3.8-flash-next

These are the two text models with a worker online in
/v1/status/models on 2026-11-13. Both files are override-only.

Sources for the host-specific values

Value	Source
Price $0.075 / $0.30 per million tokens	/v1/pricing, 2026-11-13
effort = ["low", "medium"]	The API passes low and medium on, maps none/minimal to low and drops every other value (normalize_reasoning_effort in grid_api/routers/_passthrough.py)
reasoning_field = "reasoning_content"	Live requests; an assistant turn replayed with reasoning_content is accepted by both models
structured_output = true	Live requests with response_format: json_schema on both models
limit.output = 32_768	MAX_OUTPUT_TOKENS in the same file; larger values get HTTP 422
limit.context = 40_000	The API refuses a request above 200,000 characters with HTTP 413 (MAX_REQUEST_CHARS in grid_api/routers/openai.py), whatever the model holds. 40,000 tokens stayed under that in live agent sessions; 45,000 did not always
Qwen input text, PNG, JPEG, WebP	Live requests with data URLs. Video and the other image types of the lab file were not tested and are left out

Checks

• bun validate passes.
• bun run v1:check reports only additions: the provider aipg and its two models.

🤖 Generated with Claude Code

[github-actions]

No actionable findings.