feat: add MiniMax models to the catalog

providers/model_catalog.json

@@ -198,6 +198,94 @@
           "open_weights": true
         }
       ]
+    },
+    "minimax": {
+      "latest_planning": "MiniMax-M3",
+      "latest_development": "MiniMax-M3",
+      "latest_fast": "MiniMax-M3",
+      "tier_fallback": {
+        "development": "planning",
+        "fast": "planning"
+      },
+      "notes": "MiniMax-M3 is used for all tiers; MiniMax-M2.7 (tier: legacy) is catalog-only and not a latest_* default.",
+      "endpoints": [
+        {
+          "region": "global_en",
+          "openai_base_url": "https://api.minimax.io/v1",
+          "anthropic_base_url": "https://api.minimax.io/anthropic",
+          "docs_root": "https://platform.minimax.io/docs"
+        },
+        {
+          "region": "cn_zh",
+          "openai_base_url": "https://api.minimaxi.com/v1",
+          "anthropic_base_url": "https://api.minimaxi.com/anthropic",
+          "docs_root": "https://platform.minimaxi.com/docs"
+        }
+      ],
+      "models": [
+        {
+          "id": "MiniMax-M3",
+          "tier": "planning",
+          "context_window": 1000000,
+          "max_output": 524288,
+          "input_modalities": ["text", "image", "video"],
+          "thinking": ["adaptive", "disabled"],
+          "pricing_usd_per_million_tokens": {
+            "input": 0.6,
+            "output": 2.4,
+            "cache_read": 0.12,
+            "cache_write": null
+          },
+          "pricing_tiers_usd_per_million_tokens": [
+            {
+              "service_tier": "standard",
+              "input_tokens_lte": 512000,
+              "input": 0.3,
+              "output": 1.2,
+              "cache_read": 0.06,
+              "cache_write": null
+            },
+            {
+              "service_tier": "standard",
+              "input_tokens_gt": 512000,
+              "input": 0.6,
+              "output": 2.4,
+              "cache_read": 0.12,
+              "cache_write": null
+            },
+            {
+              "service_tier": "priority",
+              "input_tokens_lte": 512000,
+              "input": 0.45,
+              "output": 1.8,
+              "cache_read": 0.09,
+              "cache_write": null
+            },
+            {
+              "service_tier": "priority",
+              "input_tokens_gt": 512000,
+              "input": 0.9,
+              "output": 3.6,
+              "cache_read": 0.18,
+              "cache_write": null
+            }
+          ]
+        },
+        {
+          "id": "MiniMax-M2.7",
+          "tier": "legacy",
+          "context_window": 204800,
+          "max_output": 204800,
+          "input_modalities": ["text"],
+          "thinking": ["always_on"],
+          "pricing_usd_per_million_tokens": {
+            "input": 0.3,
+            "output": 1.2,
+            "cache_read": 0.06,
+            "cache_write": 0.375
+          }
+        }
+      ]
     }
   }
 }

tests/run-all-tests.sh

@@ -465,6 +465,9 @@ run_test "Emit JSON Escape (C0 control chars + UTF-8)" "$SCRIPT_DIR/test-emit-js
 # against CODEX_KNOWN_MODELS. Regression guard for the silent-downgrade bug.
 run_test "Codex Model Trusted (LOKI_CODEX_MODEL verbatim)" "$SCRIPT_DIR/test-codex-model-trusted.sh"
 
+# MiniMax catalog metadata and user-configurable Anthropic/OpenAI adapter paths.
+run_test "MiniMax Model Catalog and Compatible Endpoints" "$SCRIPT_DIR/test-minimax-model-catalog.sh"
+
 # provider_invoke()/provider_invoke_with_tier() argv construction across all
 # four providers. Registered 2026-07-27: this file existed but was wired into
 # NO runner, so the layer it guards went unwatched -- which is how a hardcoded

tests/test-minimax-model-catalog.sh

@@ -0,0 +1,161 @@
+#!/usr/bin/env bash
+# Validate MiniMax model metadata and the existing compatible adapter paths.
+
+set -u
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+CATALOG="$REPO_ROOT/providers/model_catalog.json"
+PROVIDER_DOCS="$REPO_ROOT/wiki/Providers.md"
+
+PASS=0
+FAIL=0
+
+ok() { printf 'PASS: %s\n' "$1"; PASS=$((PASS + 1)); }
+bad() { printf 'FAIL: %s\n' "$1"; FAIL=$((FAIL + 1)); }
+
+if python3 - "$CATALOG" <<'PY'
+import json
+import sys
+
+with open(sys.argv[1], encoding="utf-8") as handle:
+    catalog = json.load(handle)
+
+provider = catalog["providers"]["minimax"]
+assert provider["latest_planning"] == "MiniMax-M3"
+assert provider["latest_development"] == "MiniMax-M3"
+assert provider["latest_fast"] == "MiniMax-M3"
+
+endpoints = {item["region"]: item for item in provider["endpoints"]}
+assert endpoints == {
+    "global_en": {
+        "region": "global_en",
+        "openai_base_url": "https://api.minimax.io/v1",
+        "anthropic_base_url": "https://api.minimax.io/anthropic",
+        "docs_root": "https://platform.minimax.io/docs",
+    },
+    "cn_zh": {
+        "region": "cn_zh",
+        "openai_base_url": "https://api.minimaxi.com/v1",
+        "anthropic_base_url": "https://api.minimaxi.com/anthropic",
+        "docs_root": "https://platform.minimaxi.com/docs",
+    },
+}
+assert all(item["anthropic_base_url"].endswith("/anthropic") for item in endpoints.values())
+
+models = {item["id"]: item for item in provider["models"]}
+assert list(models)[:2] == ["MiniMax-M3", "MiniMax-M2.7"]
+
+m3 = models["MiniMax-M3"]
+assert m3["context_window"] == 1_000_000
+assert m3["max_output"] == 524_288
+assert m3["input_modalities"] == ["text", "image", "video"]
+assert m3["thinking"] == ["adaptive", "disabled"]
+assert m3["pricing_usd_per_million_tokens"] == {
+    "input": 0.6,
+    "output": 2.4,
+    "cache_read": 0.12,
+    "cache_write": None,
+}
+assert m3["pricing_tiers_usd_per_million_tokens"] == [
+    {
+        "service_tier": "standard",
+        "input_tokens_lte": 512_000,
+        "input": 0.3,
+        "output": 1.2,
+        "cache_read": 0.06,
+        "cache_write": None,
+    },
+    {
+        "service_tier": "standard",
+        "input_tokens_gt": 512_000,
+        "input": 0.6,
+        "output": 2.4,
+        "cache_read": 0.12,
+        "cache_write": None,
+    },
+    {
+        "service_tier": "priority",
+        "input_tokens_lte": 512_000,
+        "input": 0.45,
+        "output": 1.8,
+        "cache_read": 0.09,
+        "cache_write": None,
+    },
+    {
+        "service_tier": "priority",
+        "input_tokens_gt": 512_000,
+        "input": 0.9,
+        "output": 3.6,
+        "cache_read": 0.18,
+        "cache_write": None,
+    },
+]
+
+m27 = models["MiniMax-M2.7"]
+assert m27["context_window"] == 204_800
+assert m27["max_output"] == 204_800
+assert m27["input_modalities"] == ["text"]
+assert m27["thinking"] == ["always_on"]
+assert m27["pricing_usd_per_million_tokens"] == {
+    "input": 0.3,
+    "output": 1.2,
+    "cache_read": 0.06,
+    "cache_write": 0.375,
+}
+PY
+then
+    ok "catalog preserves model, endpoint, context, capability, and pricing metadata"
+else
+    bad "catalog metadata does not match the supported MiniMax configuration"
+fi
+
+docs_ok=true
+for text in \
+    'https://api.minimax.io/anthropic' \
+    'https://api.minimaxi.com/anthropic' \
+    'https://api.minimax.io/v1' \
+    'https://api.minimaxi.com/v1' \
+    'ANTHROPIC_BASE_URL' \
+    'ANTHROPIC_AUTH_TOKEN' \
+    'LOKI_MODEL_OVERRIDE' \
+    'OPENAI_API_BASE' \
+    'OPENAI_API_KEY' \
+    'LOKI_AIDER_MODEL' \
+    'MiniMax-M3' \
+    'MiniMax-M2.7'; do
+    grep -Fq "$text" "$PROVIDER_DOCS" || docs_ok=false
+done
+if [ "$docs_ok" = true ]; then
+    ok "provider documentation covers both regions, protocols, and models"
+else
+    bad "provider documentation is missing required MiniMax configuration"
+fi
+
+anthropic_model="$({
+    env -i HOME="$HOME" PATH="$PATH" \
+        ANTHROPIC_BASE_URL="https://api.minimax.io/anthropic" \
+        LOKI_MODEL_OVERRIDE="MiniMax-M3" \
+        bash -c 'cd "$1"; . providers/claude.sh; resolve_model_for_tier development' _ "$REPO_ROOT"
+} 2>/dev/null)"
+if [ "$anthropic_model" = "MiniMax-M3" ]; then
+    ok "Anthropic-compatible adapter resolves the configured model"
+else
+    bad "Anthropic-compatible adapter resolved '$anthropic_model'"
+fi
+
+aider_model="$({
+    env -i HOME="$HOME" PATH="$PATH" \
+        OPENAI_API_BASE="https://api.minimax.io/v1" \
+        OPENAI_API_KEY="test-key" \
+        LOKI_AIDER_MODEL="openai/MiniMax-M2.7" \
+        bash -c 'cd "$1"; . providers/aider.sh; printf "%s" "$AIDER_DEFAULT_MODEL"' _ "$REPO_ROOT"
+} 2>/dev/null)"
+if [ "$aider_model" = "openai/MiniMax-M2.7" ]; then
+    ok "OpenAI-compatible adapter resolves the configured model"
+else
+    bad "OpenAI-compatible adapter resolved '$aider_model'"
+fi
+
+printf '\nTotal: %s  Passed: %s  Failed: %s\n' "$((PASS + FAIL))" "$PASS" "$FAIL"
+[ "$FAIL" -eq 0 ]

wiki/Providers.md

@@ -22,6 +22,48 @@ Loki Mode supports four active AI providers with different capability levels, pl
 
 ---
 
+## MiniMax API compatibility
+
+The `minimax` model catalog entry records API and model metadata. It is not a fifth value for `--provider`; use an existing CLI adapter with the compatible API protocol.
+
+### Anthropic-compatible API
+
+Choose the Base URL for the account region. Keep `/anthropic` as the end of the Base URL; the client appends `/v1/messages` when it sends a request.
+
+| Region | Base URL |
+|--------|----------|
+| Global | `https://api.minimax.io/anthropic` |
+| China | `https://api.minimaxi.com/anthropic` |
+
+```bash
+export ANTHROPIC_BASE_URL="https://api.minimax.io/anthropic"
+export ANTHROPIC_AUTH_TOKEN="<MINIMAX_API_KEY>"
+export LOKI_MODEL_OVERRIDE="MiniMax-M3"
+loki start ./prd.md --provider claude
+```
+
+Set `ANTHROPIC_BASE_URL` to the China Base URL when required. Change `LOKI_MODEL_OVERRIDE` to `MiniMax-M2.7` to use that model.
+
+### OpenAI-compatible API
+
+Choose the Base URL for the account region. Keep `/v1` as the end of the Base URL; the client appends `/chat/completions` when it sends a request.
+
+| Region | Base URL |
+|--------|----------|
+| Global | `https://api.minimax.io/v1` |
+| China | `https://api.minimaxi.com/v1` |
+
+```bash
+export OPENAI_API_BASE="https://api.minimax.io/v1"
+export OPENAI_API_KEY="<MINIMAX_API_KEY>"
+export LOKI_AIDER_MODEL="openai/MiniMax-M3"
+loki start ./prd.md --provider aider
+```
+
+Set `OPENAI_API_BASE` to the China Base URL when required. Change the model suffix to `MiniMax-M2.7` to use that model.
+
+---
+
 ## Provider precedence (v7.7.2)
 
 When multiple sources specify a provider, Loki picks the first match in this order (highest wins):