← Về thư mục
📄 / / proc / 45 / cwd / root / .hermes / skills-backup-20260914-152439 / devops / 9router-ops / references / profile-model-mapping.md

Profile → Model Pool Mapping (Canonical Reference)

Architecture: Each Hermes profile → one 9Router combo named after the profile. Model overlap across combos is INTENTIONAL and the preferred way to share models. Shared/bridge combos are UNNECESSARY and add complexity — avoid.

Priority Order

r-and-d (highest) > it-ai > grill-qa > reviewer > writers > default > bd-mkt (lowest)

Top-tier models (qwen3.7-max, qwen3.5-397b-a17b, Claude Opus, etc.) must NOT appear in default/bd-mkt combos.


1. r-and-d (research, long context, synthesis)

Antigravity

Codex

Gemini

Alibaba


2. it-ai (code, debug, systems)

Codex (ALL: supports code reasoning + review)

OpenCode / OpenRouter

Alibaba code


3. grill-qa (critique, QA, chain-of-thought)

Antigravity

Codex (review variants only)

Alibaba thinking


4. reviewer (final judgment, cross-check, synthesis)

Alibaba

Codex review

Gemini


5. writers (drafting, synthesis, Vietnamese NLP)

Alibaba

Gemini / Codex light


6. default (CEO daily Telegram, fast, cheap)


7. bd-mkt (marketing strategy, vision-capable, light)


Model Overlap Examples (by design)


Known Invalid Model IDs (as of 2026-07-20)

These were in early drafts but are NOT returned by Alibaba /models endpoint: - openai-compatible-alibaba/qwen-flash-2025-07-28 - openai-compatible-alibaba/qwen-plus-2025-07-28 - openai-compatible-alibaba/qwen3-coder-flash-2025-07-28 - openai-compatible-alibaba/qwen3-vl-32b-thinking - openrouter/qwen/qwen3-coder:free (not in active 9Router provider locks) Also: line 148 in the original doc had an unclosed backtick on qwen3-vl-30b-a3b-thinking.