MiniMax-M3 — Production Wiring (empirical)
The M3 companion to ../minimax/SKILL.md (M2.7). Every claim here was
live-probed 2026-06-01 (fast subset re-verified 2026-06-23) on the Plus-High-Speed key. Full evidence + copy-paste snippets:
../../references/M3-EMPIRICAL.md.
Self-Evolving Skill: improves through use. If a flag stopped working, a limit moved, or
the docs caught up with reality — fix this file + references/M3-EMPIRICAL.md immediately,
don't defer. Re-verify with the scripts below before changing a documented fact.
The one rule: default to reasoning_split: true
M3 still emits <think>…</think> inside content by default (same footgun as M2.7).
Setting reasoning_split: true moves the reasoning into a separate reasoning_content /
reasoning_details field and leaves content clean — no regex stripping. This is the
chosen default profile for everything migrating off M2.7-highspeed.
body = {
"model": "MiniMax-M3",
"messages": messages,
"max_tokens": 4096, # >= 1024 — thinking consumes budget before visible content
"temperature": 0.2,
"reasoning_split": True, # clean content; reasoning in reasoning_content/_details
}
answer = resp["choices"][0]["message"]["content"] # already clean — display directly
Need M2.7-highspeed-class speed on short/simple tasks? Add "reasoning": "disabled"
(≈2× fewer tokens, ≈2× faster) — and keep the M2.7 <think> strip as a safety net, since
"disabled" shortens but doesn't always remove the block. Keep thinking ON (default /
"adaptive") for hard reasoning, coding, and agentic loops.
When to use M3 vs M2.7
| Workload |
Verdict |
| Clean chat / judgment / theory / JSON |
✅ M3 + reasoning_split:true (the new default) |
| Short tagging / classification, latency-sensitive |
✅ M3 + reasoning:"disabled", or stay on plain MiniMax-M2.7 |
| Vision (OCR, charts, screenshots) |
✅ M3 only — M2.7 is text-only; M3 reads images correctly |
| Structured JSON |
✅ M3 (response_format accepted) + reasoning_split + defensive parse |
| Long context (input up to ~1M) |
✅ input accepts to ~1M, but reliable retrieval ≤ ~256K (400K now misses); 1M prefill ~235 s |
| Raw math / QP / risk on realistic N |
❌ still route to Python (the M2.7 saturation guidance carries over) |
| Final deployable code |
⚠️ scaffold-only; sandbox-validate (unchanged from M2.7) |
Hard limits & gotchas (live-verified)
- Input context ≈ 1,000,000 tokens (re-verified 2026-06-23). Accepts to ~1,000,180; 1,048,576 →
context window exceeds limit. The docs' 1M now holds (was ~512K on 2026-06-01) — but 1M
prefill is ~235 s and reliable needle retrieval is ≤ ~256K (400K misses 2/2). Operate at
≤ ~256K for retrieval-critical work.
- Output
max_tokens ≤ 524,288. > 524288 → invalid params … does not support max tokens > 524288 (raised from 512,000 — re-verified 2026-06-23).
n > 1 silently dropped. Was a hard 2013 rejection; now accepted-but-ignored — the response still carries exactly one choice. No true multi-sampling (re-verified 2026-06-23).response_format accepted but not a hard JSON guarantee — M3 may still wrap with <think>
/ ```json fences / a trailing note. Pair with reasoning_split + try/except json.loads.tool_choice forced did NOT compel a call in the trivial-prompt probe — re-test with a
tool-relevant prompt before relying on forced tool calls.- No
MiniMax-M3-highspeed on this key (2013 unknown model) despite docs.
- All the M2.7 defensive snippets (
<think> strip, base_resp rate-limit retry, cached-token
reader) in ../minimax/SKILL.md apply unchanged.
Re-verify / detect drift
Scripts live at the plugin source checkout (scripts/ is stripped from the runtime cache),
run from ~/eon/cc-skills/plugins/minimax:
export MINIMAX_API_KEY=... # or rely on the 1Password op-path default (see `bun scripts/m3-cli.ts verify --help`)
bun scripts/m3-cli.ts verify # fast drift check vs locked snapshot (exit 0/1/2)
bun scripts/m3-cli.ts probe [--out f] # full option/capability map (writes JSON; default m3_probe_results.json)
bun scripts/m3-cli.ts context-probe # input-context ceiling + needle retrieval
bun scripts/m3-cli.ts bench # speed/quality: default thinking vs reasoning:"disabled"
./scripts/minimax-check-upgrade # catalog drift (lock includes MiniMax-M3)
Locked invariants: ../../references/fixtures/m3-capabilities-locked-2026-06-23.json.
Schedule m3-cli.ts verify + minimax-check-upgrade (launchd template in templates/) to catch the
day MiniMax ships M3-highspeed, opens up 1M context, or changes a limit.
Post-Execution Reflection
- Locate yourself. — Confirm this is the canonical
skills/m3/SKILL.md before editing.
- What failed? — A flag that worked now errors, or vice-versa → fix here + M3-EMPIRICAL.md.
- What drifted? —
m3-cli verify flagged an invariant change → review, then bump the locked snapshot.
- Log it. — Append to the Evolution log in
references/M3-EMPIRICAL.md with trigger + fix + evidence.
1---2name: m33description: Production wiring for the MiniMax-M3 model — empirically verified flags, capabilities, and limits (thinking control via reasoning_split, native vision, response_format, ~1M input ceiling, 524K output cap, n=1, docs-vs-reality discrepancies). Use when wiring or tuning MiniMax-M3, choosing M3 vs M2.7/-highspeed, switching a service off M2.7-highspeed onto M3, getting clean output without <think>, or asking what M3 supports / how big its context is. TRIGGERS - MiniMax M3, MiniMax-M3, M3 model, switch to M3, reasoning_split, M3 context length, M3 vision, M3 options, get the most out of M3.4---5
6# MiniMax-M3 — Production Wiring (empirical)
7
8The M3 companion to [`../minimax/SKILL.md`](../minimax/SKILL.md) (M2.7). Every claim here was
9**live-probed 2026-06-01** (fast subset re-verified 2026-06-23) on the Plus-High-Speed key. Full evidence + copy-paste snippets:
10[`../../references/M3-EMPIRICAL.md`](../../references/M3-EMPIRICAL.md).
11
12> **Self-Evolving Skill**: improves through use. If a flag stopped working, a limit moved, or
13> the docs caught up with reality — fix this file + `references/M3-EMPIRICAL.md` immediately,
14> don't defer. Re-verify with the scripts below before changing a documented fact.
15
16---
17
18## The one rule: default to `reasoning_split: true`
19
20M3 still emits `<think>…</think>` inside `content` **by default** (same footgun as M2.7).
21Setting `reasoning_split: true` moves the reasoning into a separate `reasoning_content` /
22`reasoning_details` field and leaves `content` **clean** — no regex stripping. This is the
23chosen default profile for everything migrating off M2.7-highspeed.
24
25```python
26body = {
27 "model": "MiniMax-M3",
28 "messages": messages,
29 "max_tokens": 4096, # >= 1024 — thinking consumes budget before visible content
30 "temperature": 0.2,
31 "reasoning_split": True, # clean content; reasoning in reasoning_content/_details
32}
33answer = resp["choices"][0]["message"]["content"] # already clean — display directly
34```
35
36Need M2.7-highspeed-class **speed** on short/simple tasks? Add `"reasoning": "disabled"`
37(≈2× fewer tokens, ≈2× faster) — and keep the M2.7 `<think>` strip as a safety net, since
38`"disabled"` shortens but doesn't always remove the block. Keep thinking ON (default /
39`"adaptive"`) for hard reasoning, coding, and agentic loops.
40
41---
42
43## When to use M3 vs M2.7
44
45| Workload | Verdict |
46| ------------------------------------------------- | ------------------------------------------------------------------------------------------------ |
47| Clean chat / judgment / theory / JSON | ✅ M3 + `reasoning_split:true` (the new default) |
48| Short tagging / classification, latency-sensitive | ✅ M3 + `reasoning:"disabled"`, **or** stay on plain `MiniMax-M2.7` |
49| **Vision** (OCR, charts, screenshots) | ✅ **M3 only** — M2.7 is text-only; M3 reads images correctly |
50| Structured JSON | ✅ M3 (`response_format` accepted) + `reasoning_split` + defensive parse |
51| Long context (input up to ~1M) | ✅ input accepts to ~1M, but **reliable retrieval ≤ ~256K** (400K now misses); 1M prefill ~235 s |
52| Raw math / QP / risk on realistic N | ❌ still route to Python (the M2.7 saturation guidance carries over) |
53| Final deployable code | ⚠️ scaffold-only; sandbox-validate (unchanged from M2.7) |
54
55---
56
57## Hard limits & gotchas (live-verified)
58
59- **Input context ≈ 1,000,000 tokens (re-verified 2026-06-23).** Accepts to ~1,000,180; 1,048,576 →
60 `context window exceeds limit`. The docs' 1M now holds (was ~512K on 2026-06-01) — **but** 1M
61 prefill is ~235 s and **reliable needle retrieval is ≤ ~256K** (400K misses 2/2). Operate at
62 ≤ ~256K for retrieval-critical work.
63- **Output `max_tokens` ≤ 524,288.** > 524288 → `invalid params … does not support max tokens > 524288` (raised from 512,000 — re-verified 2026-06-23).
64- **`n > 1` silently dropped.** Was a hard `2013` rejection; now accepted-but-ignored — the response still carries exactly one choice. No true multi-sampling (re-verified 2026-06-23).
65- **`response_format` accepted but not a hard JSON guarantee** — M3 may still wrap with `<think>`
66 / ` ```json ` fences / a trailing note. Pair with `reasoning_split` + try/except `json.loads`.
67- **`tool_choice` forced did NOT compel a call** in the trivial-prompt probe — re-test with a
68 tool-relevant prompt before relying on forced tool calls.
69- **No `MiniMax-M3-highspeed`** on this key (`2013 unknown model`) despite docs.
70- All the M2.7 defensive snippets (`<think>` strip, `base_resp` rate-limit retry, cached-token
71 reader) in [`../minimax/SKILL.md`](../minimax/SKILL.md) apply unchanged.
72
73---
74
75## Re-verify / detect drift
76
77Scripts live at the plugin **source** checkout (`scripts/` is stripped from the runtime cache),
78run from `~/eon/cc-skills/plugins/minimax`:
79
80```bash
81export MINIMAX_API_KEY=... # or rely on the 1Password op-path default (see `bun scripts/m3-cli.ts verify --help`)
82
83bun scripts/m3-cli.ts verify # fast drift check vs locked snapshot (exit 0/1/2)
84bun scripts/m3-cli.ts probe [--out f] # full option/capability map (writes JSON; default m3_probe_results.json)
85bun scripts/m3-cli.ts context-probe # input-context ceiling + needle retrieval
86bun scripts/m3-cli.ts bench # speed/quality: default thinking vs reasoning:"disabled"
87./scripts/minimax-check-upgrade # catalog drift (lock includes MiniMax-M3)
88```
89
90Locked invariants: [`../../references/fixtures/m3-capabilities-locked-2026-06-23.json`](../../references/fixtures/m3-capabilities-locked-2026-06-23.json).
91Schedule `m3-cli.ts verify` + `minimax-check-upgrade` (launchd template in `templates/`) to catch the
92day MiniMax ships `M3-highspeed`, opens up 1M context, or changes a limit.
93
94## Post-Execution Reflection
95
960. **Locate yourself.** — Confirm this is the canonical `skills/m3/SKILL.md` before editing.
971. **What failed?** — A flag that worked now errors, or vice-versa → fix here + M3-EMPIRICAL.md.
982. **What drifted?** — `m3-cli verify` flagged an invariant change → review, then bump the locked snapshot.
993. **Log it.** — Append to the Evolution log in `references/M3-EMPIRICAL.md` with trigger + fix + evidence.
terrylica