fix(minimax): align default_aux_model with M3 frontier on minimax + minimax-cn by Fearvox · Pull Request #37664 · NousResearch/hermes-agent

Fearvox · 2026-06-02T21:40:39Z

Problem

plugins/model-providers/minimax/__init__.py still advertises stale aux models for all three MiniMax profiles, even after M3 (the new frontier, released 2026-06-01) became the catalog top entry in _PROVIDER_MODELS["minimax"] and the recommended model.default for a Token-Plan install.

Profile	Old `default_aux_model`	Why it's wrong
`minimax`	`MiniMax-M2.7`	Pre-M3 era. Catalog and `model.default` are now M3.
`minimax-cn`	`MiniMax-M2.7`	Same.
`minimax-oauth`	`MiniMax-M2.7-highspeed`	The 2x-cost regression from #4082 that PR #6082 collapsed to plain `MiniMax-M2.7` for the two direct API providers — but the OAuth path was missed.

agent/auxiliary_client.py also still has a legacy fallback dict entry for each of the three providers, which _get_aux_model_for_provider() only consults when the profile import fails or the profile has no aux model. With the profile set, the dict entries are dead code and a drift hazard.

This is the "profile layer" half of #36196's M3-support work; #36212 covers the catalog half (_PROVIDER_MODELS["minimax"]).

Changes

plugins/model-providers/minimax/__init__.py
- minimax: default_aux_model M2.7 → M3
- minimax-cn: default_aux_model M2.7 → M3
- minimax-oauth: default_aux_model M2.7-highspeed → M2.7 (M3 is not on the OAuth / Coding Plan tier; the highspeed variant was the 2x-cost regression from [Bug]: Minimax Default Auxiallry Model MiniMax-M2.7-highspeed is twice the price at API rates #4082 that PR fix(minimax): correct context lengths, model catalog, thinking guard, aux model, and config base_url #6082 collapsed to plain M2.7 for the direct API providers but missed OAuth)
agent/auxiliary_client.py
- Drop the three minimax / minimax-oauth / minimax-cn entries from _API_KEY_PROVIDER_AUX_MODELS_FALLBACK. Dead code once the profile is set, and a drift hazard. Mirrors the cleanup in 773a0fa for deepseek.
tests/agent/test_minimax_provider.py
- Update TestMinimaxAuxModel assertions from MiniMax-M2.7 to MiniMax-M3 (intent — "standard, not highspeed" — is unchanged; the pin value is).
tests/plugins/model_providers/test_minimax_profile.py (new file, mirrors test_deepseek_profile.py)
- TestMinimaxAuxModelM3: parametrize over all three profiles, assert default_aux_model matches the expected per-provider value (M3 for direct API, M2.7 for OAuth). Assert _get_aux_model_for_provider() returns the same value the profile advertises (no drift between the consumer API and the profile layer).
- TestMinimaxAuxModelNotHighspeed: regression guard against the highspeed 2x-cost path coming back. This was the core of [Bug]: Minimax Default Auxiallry Model MiniMax-M2.7-highspeed is twice the price at API rates #4082.

Verification

$ python3 -m pytest tests/agent/test_minimax_provider.py tests/plugins/model_providers/ tests/test_minimax_model_validation.py tests/test_minimax_oauth.py tests/agent/test_minimax_auxiliary_url.py
============================= 100 passed in 3.27s ==============================

Live POST https://api.minimax.io/anthropic/v1/messages with model: "MiniMax-M3" against MINIMAX_API_KEY returns 200 with a thinking-block response, confirming M3 is the right model ID for the aux default.

Closes [Feature]: MinimaxM3 model support #36196 (in spirit — the catalog half is feat(models): add MiniMax-M3 + highspeed variants to direct minimax provider list #36212, this PR covers the profile half)
Related: [Bug]: Minimax Default Auxiallry Model MiniMax-M2.7-highspeed is twice the price at API rates #4082 (M2.7-highspeed 2x-cost), fix(minimax): correct context lengths, model catalog, thinking guard, aux model, and config base_url #6082 (previous M2.7-highspeed → M2.7 fix that missed OAuth + the auxiliary_client.py fallback dict)
Pattern: 773a0fa (same profile-layer fix for deepseek)

Copilot

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Updates MiniMax provider profiles and tests to align the default auxiliary model with the newer M3 model (while keeping OAuth on M2.7) and to prevent regressions back to “-highspeed”.

Changes:

Update MiniMax provider profiles to default default_aux_model to MiniMax-M3 (and MiniMax-M2.7 for OAuth).
Add/adjust tests to pin expected aux model defaults and ensure profiles + consumer API stay in sync.
Remove MiniMax entries from the legacy aux-model fallback map.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
tests/plugins/model_providers/test_minimax_profile.py	Adds profile-level tests pinning expected `default_aux_model` values and ensuring consumer API resolution matches.
tests/agent/test_minimax_provider.py	Updates aux-model expectations to M3 for `minimax`/`minimax-cn` and documents the split for OAuth.
plugins/model-providers/minimax/init.py	Changes provider profile defaults to M3 for direct API and to M2.7 for OAuth.
agent/auxiliary_client.py	Removes legacy fallback mappings for MiniMax, pushing resolution entirely through provider profiles.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

    "stepfun": "step-3.5-flash",
    "kimi-coding-cn": "kimi-k2-turbo-preview",
    "gmi": "google/gemini-3.1-flash-lite-preview",
-    "minimax": "MiniMax-M2.7",
-    "minimax-oauth": "MiniMax-M2.7-highspeed",
-    "minimax-cn": "MiniMax-M2.7",
    "anthropic": "claude-haiku-4-5-20251001",
    "opencode-zen": "gemini-3-flash",
    "opencode-go": "glm-5",


    def test_minimax_aux_is_standard(self):
        from agent.auxiliary_client import _get_aux_model_for_provider
-        assert _get_aux_model_for_provider("minimax") == "MiniMax-M2.7"
-        assert _get_aux_model_for_provider("minimax-cn") == "MiniMax-M2.7"
+        assert _get_aux_model_for_provider("minimax") == "MiniMax-M3"
+        assert _get_aux_model_for_provider("minimax-cn") == "MiniMax-M3"


+    The catalog top entry is ``MiniMax-M3`` (hermes_cli/models.py:298) and
+    the user-facing ``model.default`` for a Token-Plan install is M3, so


…inimax-cn The minimax / minimax-cn / minimax-oauth profiles still advertised M2.7 (and M2.7-highspeed for OAuth) as their default_aux_model, predating the M3 release (2026-06-01). The user-facing _PROVIDER_MODELS['minimax'] catalog top entry is M3, and the recommended config for a Token-Plan install now sets model.default: MiniMax-M3, so the aux default was the only remaining drift. Updates: * minimax default_aux_model: M2.7 -> M3 * minimax-cn default_aux_model: M2.7 -> M3 * minimax-oauth default_aux_model: M2.7-highspeed -> M2.7 (M3 is not on the OAuth / Coding Plan tier per platform docs as of this PR; the highspeed variant was the 2x-cost regression from NousResearch#4082 that PR NousResearch#6082 collapsed to plain M2.7 for minimax / minimax-cn but missed OAuth) * agent/auxiliary_client.py: drop the three legacy _API_KEY_PROVIDER_AUX_MODELS_FALLBACK entries for the minimax family. _get_aux_model_for_provider() reads from ProviderProfile.default_aux_model first (line 250) and only falls back to the dict when the profile has no aux model or the profile import fails. With the profile now set, the dict entries are dead code and a drift hazard. Mirrors the deepseek cleanup in 773a0fa. * tests/agent/test_minimax_provider.py: update the existing TestMinimaxAuxModel assertions from MiniMax-M2.7 to MiniMax-M3 (the intent — 'standard, not highspeed' — is unchanged; the pin value is). * tests/plugins/model_providers/test_minimax_profile.py: new file mirroring tests/plugins/model_providers/test_deepseek_profile.py. Pins each of the three profiles' default_aux_model and asserts _get_aux_model_for_provider() returns it. A second class guards against the highspeed regression coming back. Refs: - Closes NousResearch#36196 in spirit (M3 support — the catalog half of that issue is NousResearch#36212; this PR covers the profile half) - Related: NousResearch#4082 (M2.7-highspeed 2x-cost), NousResearch#6082 (previous M2.7-highspeed -> M2.7 fix that missed OAuth + the auxiliary_client.py fallback dict) - Pattern: 773a0fa (same profile-layer fix for deepseek)

…registered Two follow-ups to the M3 default-aux-model PR (NousResearch#37664): 1. AUTHOR_MAP entry: add fearvox1015@gmail.com -> Fearvox so the check-attribution CI job recognises Nolan's real contributor email. The previous run of the attribution check on NousResearch#37664 failed because the commit was authored as nolan@0xvox.com (wrong local git config) which isn't in AUTHOR_MAP. The commit itself is now re-authored to fearvox1015@gmail.com so both the per-commit check and the AUTHOR_MAP lookup pass. 2. tests/hermes_cli/test_api_key_providers.py::TestMinimaxOAuthProvider ::test_minimax_oauth_aux_model_registered was pinning the aux model in the legacy _API_KEY_PROVIDER_AUX_MODELS dict, which the PR correctly removed (mirrors the deepseek cleanup in 773a0fa). The test now asserts the new world order: the aux model comes from ProviderProfile.default_aux_model on the minimax-oauth profile, not the fallback dict. This is the same pattern that the profile-layer deepseek fix introduced.

Fearvox · 2026-06-02T22:22:15Z

Pushed a fix for the two CI failures:

test (2) failure (this PR's only test failure): tests/hermes_cli/test_api_key_providers.py::TestMinimaxOAuthProvider::test_minimax_oauth_aux_model_registered was pinning the aux model in the legacy _API_KEY_PROVIDER_AUX_MODELS dict, which this PR correctly removed (mirrors the deepseek cleanup in 773a0fa). Updated the test to assert the new world order: the aux model comes from ProviderProfile.default_aux_model on the minimax-oauth profile, not the fallback dict. Same pattern that the profile-layer deepseek fix introduced.
check-attribution failure (both PRs): the commit author email was nolan@0xvox.com (a local git config mistake on this machine — should have been fearvox1015@gmail.com). Re-authored the commit and added the fearvox1015@gmail.com → Fearvox entry to scripts/release.py AUTHOR_MAP so the check passes for any future commits authored from this email.

Diff stat now:

scripts/release.py +1 line (AUTHOR_MAP entry)
tests/hermes_cli/test_api_key_providers.py updated test assertion

$ python3 -m pytest tests/hermes_cli/test_api_key_providers.py -v
============================= 165 passed in 14.37s =============================

Three Copilot inline review comments on NousResearch#37664, two worth landing in a polish pass before merge: 1. auxiliary_client.py:270 — Copilot suggested keeping the minimax-* entries in _API_KEY_PROVIDER_AUX_MODELS_FALLBACK as a safety net for environments where the profile-based resolution can't import or run plugin discovery. **Declined.** The deepseek precedent (commit 773a0fa) explicitly removed deepseek from the same dict for the same reason — the profile layer is the source of truth and the dict is a legacy pre-profiles-system fallback. We do not want to fragment the codebase by provider: either the profile layer is authoritative or the dict is. The minimax PR picks profile (matching deepseek) and the dict stays cleaned up. The risk Copilot raises is real but theoretical — plugin discovery runs at import time of the providers module, which is the first thing any modern Hermes entrypoint imports. 2. tests/agent/test_minimax_provider.py:162 — Copilot flagged that the test class relies on _get_aux_model_for_provider() resolving via provider profiles but doesn't explicitly trigger plugin discovery. **Fixed.** Added 'import model_tools # noqa: F401' at the top of both test_minimax_aux_is_standard and test_minimax_aux_not_highspeed. The fixtures in the parallel test_minimax_profile.py already did this; the legacy test in test_minimax_provider.py was order-dependent and would silently break if anyone reorganised the test ordering. Pinned the dependency explicitly so the test is order-independent. 3. tests/plugins/model_providers/test_minimax_profile.py:46 — Copilot flagged that the docstring referenced a hard-coded line number 'hermes_cli/models.py:298' that would go stale. **Fixed.** Replaced with the symbol reference 'hermes_cli.models._PROVIDER_MODELS[\'minimax\']' which is stable under file edits and grep-friendly. The new docstring also reads more naturally — readers don't have to look up 'what's at line 298' to follow the reasoning. All 221 minimax-related tests still pass.

Fearvox · 2026-06-02T22:39:13Z

Addressed the worth-doing Copilot review comments in a polish commit:

#37664 (minimax M3):

✅ Copilot Support passing morph snapshot id #2 — Added import model_tools to tests/agent/test_minimax_provider.py so the legacy aux-model test class triggers plugin discovery explicitly (was order-dependent before).
✅ Copilot Architecture planning #3 — Replaced hard-coded hermes_cli/models.py:298 line number with the symbol _PROVIDER_MODELS['minimax'] in the new test file's docstring.
⏭️ Copilot Terminal tool #1 — Declined the fallback-dict safety net suggestion. The deepseek precedent (commit 773a0fa) explicitly removed deepseek from the same dict for the same reasoning; we want consistency, not per-provider fragmentation. Explained in the commit body.

#37679 (fd-leak):

✅ Copilot Terminal tool #1, Support passing morph snapshot id #2 — Updated _dispose_unused_adapter signature to BasePlatformAdapter | None.
✅ Copilot Architecture planning #3 — Added an explicit comment explaining that asyncio.CancelledError inherits from BaseException (not Exception) on Python 3.8+, so the existing except Exception does not swallow task cancellation. Hermes requires Python 3.13 per pyproject.toml. We do not re-raise explicitly because the watcher loop intentionally treats dispose failures as best-effort.
✅ Copilot Fix VM instance sharing across tasks #6 — Replaced bare pytest.raises(Exception) with pytest.raises(sqlite3.ProgrammingError) so the test only passes when the SQLite conn is actually closed, not on any unrelated exception.
✅ Copilot some cleanups #7 — Tightened adapter._disconnect_calls assertion from >= 1 to == 1. Catches the hypothetical 'watcher disposes the same adapter twice' regression that >= would have missed.
⏭️ Copilot Fix terminal interactivity #4 — Declined setting _response_store = None after close. The current code is already safe (close() is idempotent on a closed sqlite3 Connection) and setting to None would cascade to ~9 callers that don't have None checks, generating LSP cascade errors. The reviewer's concern is real but the cost-benefit favors the current code.
⏭️ Copilot Update snapshot #5 — Already handled. _build_adapter_with_store sets adapter.platform = Platform.API_SERVER so the read-only self.name property returns without AttributeError when the logger fires in the close-exception path.

Both branches re-pushed. Local test runs:

minimax: 221/221 pass
fd-leak: 36/36 pass

…registered Two follow-ups to the M3 default-aux-model PR (#37664): 1. AUTHOR_MAP entry: add fearvox1015@gmail.com -> Fearvox so the check-attribution CI job recognises Nolan's real contributor email. The previous run of the attribution check on #37664 failed because the commit was authored as nolan@0xvox.com (wrong local git config) which isn't in AUTHOR_MAP. The commit itself is now re-authored to fearvox1015@gmail.com so both the per-commit check and the AUTHOR_MAP lookup pass. 2. tests/hermes_cli/test_api_key_providers.py::TestMinimaxOAuthProvider ::test_minimax_oauth_aux_model_registered was pinning the aux model in the legacy _API_KEY_PROVIDER_AUX_MODELS dict, which the PR correctly removed (mirrors the deepseek cleanup in 773a0fa). The test now asserts the new world order: the aux model comes from ProviderProfile.default_aux_model on the minimax-oauth profile, not the fallback dict. This is the same pattern that the profile-layer deepseek fix introduced.

Three Copilot inline review comments on #37664, two worth landing in a polish pass before merge: 1. auxiliary_client.py:270 — Copilot suggested keeping the minimax-* entries in _API_KEY_PROVIDER_AUX_MODELS_FALLBACK as a safety net for environments where the profile-based resolution can't import or run plugin discovery. **Declined.** The deepseek precedent (commit 773a0fa) explicitly removed deepseek from the same dict for the same reason — the profile layer is the source of truth and the dict is a legacy pre-profiles-system fallback. We do not want to fragment the codebase by provider: either the profile layer is authoritative or the dict is. The minimax PR picks profile (matching deepseek) and the dict stays cleaned up. The risk Copilot raises is real but theoretical — plugin discovery runs at import time of the providers module, which is the first thing any modern Hermes entrypoint imports. 2. tests/agent/test_minimax_provider.py:162 — Copilot flagged that the test class relies on _get_aux_model_for_provider() resolving via provider profiles but doesn't explicitly trigger plugin discovery. **Fixed.** Added 'import model_tools # noqa: F401' at the top of both test_minimax_aux_is_standard and test_minimax_aux_not_highspeed. The fixtures in the parallel test_minimax_profile.py already did this; the legacy test in test_minimax_provider.py was order-dependent and would silently break if anyone reorganised the test ordering. Pinned the dependency explicitly so the test is order-independent. 3. tests/plugins/model_providers/test_minimax_profile.py:46 — Copilot flagged that the docstring referenced a hard-coded line number 'hermes_cli/models.py:298' that would go stale. **Fixed.** Replaced with the symbol reference 'hermes_cli.models._PROVIDER_MODELS[\'minimax\']' which is stable under file edits and grep-friendly. The new docstring also reads more naturally — readers don't have to look up 'what's at line 298' to follow the reasoning. All 221 minimax-related tests still pass.

teknium1 · 2026-06-04T12:54:03Z

Merged via PR #39024 onto current main. Your three commits were cherry-picked with your authorship preserved in git log (commits 3d1d0a49f, b531b5d12, fa8e2f935).

The AUTHOR_MAP line in your second commit auto-resolved since fearvox1015@gmail.com was already on main, but the substantive test fix from that commit landed. M3 is now the aux default for minimax / minimax-cn and M2.7 for minimax-oauth, with the dead fallback-dict entries dropped. Thanks for the clean, well-tested fix.