feat: upgrade MiniMax default model to M2.7

PR Bot · PR Bot · commit 2a48c29896ac · 2026-03-18T16:49:34.000+08:00
- Add MiniMax-M2.7 and MiniMax-M2.7-highspeed to ModelName type
- Set MiniMax-M2.7 as default model for all agents (base2-free,
  editor-lite, code-reviewer-lite, librarian, tmux-cli)
- Update FREE_MODE_AGENT_MODELS allowlist to M2.7
- Keep all previous models (M2.5) as available alternatives
- Update documentation (what-models, modes, FAQ, how-does-it-work,
  freebuff README/SPEC)
- Update related unit tests
diff --git a/.agents/types/agent-definition.ts b/.agents/types/agent-definition.ts
@@ -429,6 +429,8 @@ export type ModelName =
   | 'z-ai/glm-4.7:nitro'
   | 'z-ai/glm-4.7-flash'
   | 'z-ai/glm-4.7-flash:nitro'
+  | 'minimax/minimax-m2.7'
+  | 'minimax/minimax-m2.7-highspeed'
   | 'minimax/minimax-m2.5'
   | (string & {})
 
diff --git a/agents/__tests__/editor.test.ts b/agents/__tests__/editor.test.ts
@@ -64,7 +64,7 @@ describe('editor agent', () => {
 
     test('creates minimax editor', () => {
       const minimaxEditor = createCodeEditor({ model: 'minimax' })
-      expect(minimaxEditor.model).toBe('minimax/minimax-m2.5')
+      expect(minimaxEditor.model).toBe('minimax/minimax-m2.7')
     })
 
     test('gpt-5 editor does not include think tags in instructions', () => {
diff --git a/agents/base2/base2.ts b/agents/base2/base2.ts
@@ -28,7 +28,7 @@ export function createBase2(
 
   return {
     publisher,
-    model: isFree ? 'minimax/minimax-m2.5' : 'anthropic/claude-opus-4.6',
+    model: isFree ? 'minimax/minimax-m2.7' : 'anthropic/claude-opus-4.6',
     providerOptions: isFree ? {
       data_collection: 'deny',
     } : {
diff --git a/agents/editor/editor.ts b/agents/editor/editor.ts
@@ -13,7 +13,7 @@ export const createCodeEditor = (options: {
       options.model === 'gpt-5'
         ? 'openai/gpt-5.1'
         : options.model === 'minimax'
-          ? 'minimax/minimax-m2.5'
+          ? 'minimax/minimax-m2.7'
           : 'anthropic/claude-opus-4.6',
     ...(options.model === 'opus' && {
       providerOptions: {
diff --git a/agents/librarian/librarian.ts b/agents/librarian/librarian.ts
@@ -9,7 +9,7 @@ const librarian: AgentDefinition = {
   id: 'librarian',
   publisher,
   displayName: 'Librarian',
-  model: 'minimax/minimax-m2.5',
+  model: 'minimax/minimax-m2.7',
 
   spawnerPrompt:
     'Spawn the librarian agent to shallow-clone a GitHub repository into /tmp and answer questions about its code, structure, or documentation. The agent returns structured output with `answer`, `relevantFiles` (absolute paths in the cloned repo), and `cloneDir`. You can use `run_terminal_command` with `cat` to read the returned `relevantFiles` paths. Clean up `cloneDir` with `rm -rf` when done.',
diff --git a/agents/reviewer/code-reviewer-lite.ts b/agents/reviewer/code-reviewer-lite.ts
@@ -5,7 +5,7 @@ import { createReviewer } from './code-reviewer'
 const definition: SecretAgentDefinition = {
   id: 'code-reviewer-lite',
   publisher,
-  ...createReviewer('minimax/minimax-m2.5'),
+  ...createReviewer('minimax/minimax-m2.7'),
 }
 
 export default definition
diff --git a/agents/tmux-cli.ts b/agents/tmux-cli.ts
@@ -71,7 +71,7 @@ const outputSchema = {
 const definition: AgentDefinition = {
   id: 'tmux-cli',
   displayName: 'Tmux CLI Agent',
-  model: 'minimax/minimax-m2.5',
+  model: 'minimax/minimax-m2.7',
   // Provider options are tightly coupled to the model choice above.
   // If you change the model, update these accordingly.
   providerOptions: {
diff --git a/agents/types/agent-definition.ts b/agents/types/agent-definition.ts
@@ -429,6 +429,8 @@ export type ModelName =
   | 'z-ai/glm-4.7:nitro'
   | 'z-ai/glm-4.7-flash'
   | 'z-ai/glm-4.7-flash:nitro'
+  | 'minimax/minimax-m2.7'
+  | 'minimax/minimax-m2.7-highspeed'
   | 'minimax/minimax-m2.5'
   | (string & {})
 
diff --git a/common/src/constants/free-agents.ts b/common/src/constants/free-agents.ts
@@ -18,7 +18,7 @@ export const FREE_COST_MODE = 'free' as const
  */
 export const FREE_MODE_AGENT_MODELS: Record<string, Set<string>> = {
   // Root orchestrator
-  'base2-free': new Set(['minimax/minimax-m2.5']),
+  'base2-free': new Set(['minimax/minimax-m2.7']),
 
   // File exploration agents
   'file-picker': new Set(['google/gemini-2.5-flash-lite']),
@@ -33,10 +33,10 @@ export const FREE_MODE_AGENT_MODELS: Record<string, Set<string>> = {
   'basher': new Set(['google/gemini-3.1-flash-lite-preview']),
 
   // Editor for free mode
-  'editor-lite': new Set(['minimax/minimax-m2.5']),
+  'editor-lite': new Set(['minimax/minimax-m2.7']),
 
   // Code reviewer for free mode
-  'code-reviewer-lite': new Set(['minimax/minimax-m2.5']),
+  'code-reviewer-lite': new Set(['minimax/minimax-m2.7']),
 }
 
 /**
@@ -103,7 +103,7 @@ export function isFreeModeAllowedAgentModel(
   // Exact match first
   if (allowedModels.has(model)) return true
 
-  // OpenRouter may return dated variants (e.g. "minimax/minimax-m2.5-20260211")
+  // OpenRouter may return dated variants (e.g. "minimax/minimax-m2.7-20260318")
   // so also check if the returned model starts with any allowed model prefix.
   for (const allowed of allowedModels) {
     if (model.startsWith(allowed + '-')) return true
diff --git a/common/src/templates/initial-agents-dir/types/agent-definition.ts b/common/src/templates/initial-agents-dir/types/agent-definition.ts
@@ -429,6 +429,8 @@ export type ModelName =
   | 'z-ai/glm-4.7:nitro'
   | 'z-ai/glm-4.7-flash'
   | 'z-ai/glm-4.7-flash:nitro'
+  | 'minimax/minimax-m2.7'
+  | 'minimax/minimax-m2.7-highspeed'
   | 'minimax/minimax-m2.5'
   | (string & {})
 
diff --git a/freebuff/README.md b/freebuff/README.md
@@ -54,7 +54,7 @@ freebuff
 
 **How can it be free?** Freebuff is supported by ads shown in the CLI.
 
-**What models do you use?** MiniMax M2.5 as the main coding agent, Gemini 3.1 Flash Lite for finding files and research, and GPT-5.4 for deep thinking if you connect your ChatGPT subscription.
+**What models do you use?** MiniMax M2.7 as the main coding agent, Gemini 3.1 Flash Lite for finding files and research, and GPT-5.4 for deep thinking if you connect your ChatGPT subscription.
 
 **Are you training on my data?** No. We only use model providers that do not train on our requests. Your code stays yours.
 
diff --git a/freebuff/SPEC.md b/freebuff/SPEC.md
@@ -84,7 +84,7 @@ Freebuff only supports **FREE mode**. All mode-related features are stripped.
 | `/agent:gpt-5` | Premium agent, not available in free tier |
 | `/review` | Uses thinker-gpt under the hood |
 | `/publish` | Agent publishing not available in free tier |
-| `/image` (+ `/img`, `/attach`) | Image attachments unavailable with free model (MiniMax M2.5) |
+| `/image` (+ `/img`, `/attach`) | Image attachments unavailable with free model (MiniMax M2.7) |
 
 ### Commands to KEEP
 
diff --git a/freebuff/web/src/app/home-client.tsx b/freebuff/web/src/app/home-client.tsx
@@ -29,7 +29,7 @@ const faqs = [
   {
     question: 'What models do you use?',
     answer:
-      'MiniMax M2.5 as the main coding agent. Gemini 3.1 Flash Lite for finding files and research.\n\nConnect your ChatGPT subscription to unlock GPT-5.4 for deep thinking.',
+      'MiniMax M2.7 as the main coding agent. Gemini 3.1 Flash Lite for finding files and research.\n\nConnect your ChatGPT subscription to unlock GPT-5.4 for deep thinking.',
   },
   {
     question: 'Which countries is Freebuff available in?',
diff --git a/web/src/content/advanced/how-does-it-work.mdx b/web/src/content/advanced/how-does-it-work.mdx
@@ -25,7 +25,7 @@ The main agent ("Buffy") runs on Claude Opus 4.6. It reads your prompt, gathers
 - [**Researcher**](/publishers/codebuff/agents/researcher) (Grok 4 Fast) - web and docs lookup
 - [**Thinker**](/publishers/codebuff/agents/thinker) (GPT-5.1, Gemini 2.5 Pro) - works through hard problems
 - [**Editor**](/publishers/codebuff/agents/editor) (GPT-5.1, Claude Opus 4.6) - writes and modifies code
-- [**Reviewer**](/publishers/codebuff/agents/reviewer) (Claude Opus 4.6, MiniMax M2.5 in Free mode) - catches bugs and style issues
+- [**Reviewer**](/publishers/codebuff/agents/reviewer) (Claude Opus 4.6, MiniMax M2.7 in Free mode) - catches bugs and style issues
 - [**Basher**](/publishers/codebuff/agents/basher) (Gemini 3.1 Flash Lite) - runs terminal commands
 
 ## Best-of-N Selection (Max Mode)
diff --git a/web/src/content/advanced/what-models.mdx b/web/src/content/advanced/what-models.mdx
@@ -19,7 +19,7 @@ The main agent ("Buffy") coordinates everything:
   | Default | Opus 4.6 |
   | Plan | Opus 4.6 |
   | Max | Opus 4.6 |
-  | Free | MiniMax M2.5 |
+  | Free | MiniMax M2.7 |
 </MarkdownTable>
 
 ## Subagents
@@ -29,12 +29,12 @@ The orchestrator spawns these for specific jobs:
 <MarkdownTable>
   | Task | Models |
   |------|--------|
-  | Code editing | Claude Opus 4.6, Minimax M2.5 |
+  | Code editing | Claude Opus 4.6, MiniMax M2.7 |
   | Thinking/reasoning | Claude Opus 4.6, GPT-5.4 |
   | Code review | Claude Opus 4.6, GPT-5.4 |
   | File discovery | Gemini 3.1 Flash Lite, Gemini 2.5 Flash Lite |
   | Terminal commands | Gemini 3.1 Flash Lite |
   | Web/docs research | Gemini 3.1 Flash Lite |
 </MarkdownTable>
 
-Max mode runs multiple implementations in parallel and picks the best one. Default mode runs a single implementation pass. Free mode uses MiniMax M2.5 and includes code review support.
+Max mode runs multiple implementations in parallel and picks the best one. Default mode runs a single implementation pass. Free mode uses MiniMax M2.7 and includes code review support.
diff --git a/web/src/content/help/faq.mdx b/web/src/content/help/faq.mdx
@@ -13,7 +13,7 @@ Software development: Writing features, tests, and scripts across common languag
 
 ## What model does Codebuff use?
 
-Multiple. The orchestrator uses Claude Opus 4.6 in Default and Max modes, or MiniMax M2.5 in Free mode. Subagents are matched to their tasks: Claude Opus 4.6 for code editing, GPT-5.1 for deep reasoning, Grok 4.1 Fast for terminal commands and research, and Relace AI for fast file rewrites. Free mode includes code review support. See [What models do you use?](/docs/advanced/what-models) for the full breakdown.
+Multiple. The orchestrator uses Claude Opus 4.6 in Default and Max modes, or MiniMax M2.7 in Free mode. Subagents are matched to their tasks: Claude Opus 4.6 for code editing, GPT-5.1 for deep reasoning, Grok 4.1 Fast for terminal commands and research, and Relace AI for fast file rewrites. Free mode includes code review support. See [What models do you use?](/docs/advanced/what-models) for the full breakdown.
 
 ## Can I use my Claude Pro or Max subscription with Codebuff?
 
diff --git a/web/src/content/tips/modes.mdx b/web/src/content/tips/modes.mdx
@@ -15,7 +15,7 @@ Codebuff has four modes. Switch during a session with `Shift+Tab` or `/mode:` co
   | Default | Claude Opus 4.6 | editor | Yes |
   | Max | Claude Opus 4.6 | editor-multi-prompt | Yes |
   | Plan | Claude Opus 4.6 | None | No |
-  | Free | MiniMax M2.5 | editor-lite | No |
+  | Free | MiniMax M2.7 | editor-lite | No |
 </MarkdownTable>
 
 ## Default
@@ -60,7 +60,7 @@ Switch to this mode with `/mode:plan`.
 
 ## Free
 
-MiniMax M2.5, cheaper and faster:
+MiniMax M2.7, cheaper and faster:
 
 - Less file context gathering
 - Skips code review

Original file line number	Diff line number	Diff line change
`@@ -5,7 +5,7 @@ import { createReviewer } from './code-reviewer'`
`5`	`5`	`const definition: SecretAgentDefinition = {`
`6`	`6`	`id: 'code-reviewer-lite',`
`7`	`7`	`publisher,`
`8`		`- ...createReviewer('minimax/minimax-m2.5'),`
	`8`	`+ ...createReviewer('minimax/minimax-m2.7'),`
`9`	`9`	`}`
`10`	`10`
`11`	`11`	`export default definition`