MiniMax M3

Third-party
Zero data retention

MiniMax's M3 language model with frontier coding and agentic capabilities, a 1M token context window, and multilingual support.

Model Info
Context Window ↗	1,000,000 tokens
Terms and License	link ↗
More information	link ↗
Zero data retention	Yes
Request formats	Chat Completions, Anthropic Messages
Pricing	View pricing in the Cloudflare dashboard ↗

Usage

const response = await env.AI.run(
  'minimax/m3',
  { max_tokens: 2048, messages: [{ content: 'What is the capital of France?', role: 'user' }] },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/v1/chat/completions \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "minimax/m3",
  "max_tokens": 2048,
  "messages": [
    {
      "content": "What is the capital of France?",
      "role": "user"
    }
  ]
}'

The capital of France is **Paris**.

{
  "id": "066cf8a95614a7ebfcaa6ffd7f2cd663",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "The capital of France is **Paris**.",
        "name": "MiniMax AI",
        "audio_content": "",
        "reasoning_content": "The user is asking a simple factual question: \"What is the capital of France?\" This is straightforward - the capital of France is Paris.",
        "reasoning_details": [
          {
            "type": "reasoning.text",
            "id": "reasoning-text-1",
            "format": "MiniMax-response-v1",
            "index": 0,
            "text": "The user is asking a simple factual question: \"What is the capital of France?\" This is straightforward - the capital of France is Paris."
          }
        ]
      },
      "finish_reason": "stop"
    }
  ],
  "created": 1780336041,
  "model": "MiniMax-M3",
  "object": "chat.completion",
  "usage": {
    "total_tokens": 221,
    "prompt_tokens": 183,
    "completion_tokens": 38,
    "total_characters": 0,
    "prompt_tokens_details": {
      "cached_tokens": 114
    }
  },
  "input_sensitive": false,
  "output_sensitive": false,
  "base_resp": {
    "status_code": 0,
    "status_msg": ""
  },
  "input_sensitive_type": 0,
  "output_sensitive_type": 0,
  "output_sensitive_int": 0,
  "gatewayMetadata": {
    "keySource": "Unified"
  }
}

Examples

With System Prompt — Using a system prompt to guide behavior

const response = await env.AI.run(
  'minimax/m3',
  {
    max_tokens: 2048,
    messages: [
      {
        content:
          'You are a helpful cooking assistant. Give concise recipes with metric measurements.',
        role: 'system',
      },
      { content: 'How do I make a simple pasta aglio e olio?', role: 'user' },
    ],
    temperature: 0.7,
  },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/v1/chat/completions \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "minimax/m3",
  "max_tokens": 2048,
  "messages": [
    {
      "content": "You are a helpful cooking assistant. Give concise recipes with metric measurements.",
      "role": "system"
    },
    {
      "content": "How do I make a simple pasta aglio e olio?",
      "role": "user"
    }
  ],
  "temperature": 0.7
}'

# Pasta Aglio e Olio

A classic Italian dish that comes together in minutes with just a few ingredients.

**Serves:** 2 | **Prep:** 5 min | **Cook:** 15 min

## Ingredients
- 200 g spaghetti or linguine
- 80 ml extra virgin olive oil
- 4 garlic cloves, thinly sliced
- 1–2 small dried red chilies, crushed (or ¼ tsp chili flakes)
- Salt, for pasta water
- 15 g fresh flat-leaf parsley, chopped
- Grated Parmesan or Pecorino (optional, traditionalists skip it)

## Method
1. **Boil pasta:** Bring a large pot of well-salted water to a rolling boil. Cook the pasta until al dente (about 1 minute less than package instructions). Reserve 1 cup of pasta water before draining.
2. **Infuse oil:** While the pasta cooks, gently heat the olive oil in a wide pan over low-medium heat. Add the garlic and chilies, and cook slowly for 2–3 minutes until the garlic turns pale golden. **Don't let it brown** — it will turn bitter.
3. **Combine:** Add the drained pasta to the pan with a splash of reserved pasta water. Toss vigorously over medium heat for about 1 minute so the pasta absorbs the oil.
4. **Finish:** Add more pasta water as needed to create a light, silky sauce that coats the noodles. Stir in the parsley and season with salt.
5. **Serve** immediately, topped with cheese if using.

## Tips
- Cold olive oil + slow heat = better flavor extraction.
- The starch in the pasta water is key to a glossy, emulsified sauce.

Buon appetito! 🍝

{
  "id": "066cf8abfaba58c357c660ebfad38073",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "# Pasta Aglio e Olio\n\nA classic Italian dish that comes together in minutes with just a few ingredients.\n\n**Serves:** 2 | **Prep:** 5 min | **Cook:** 15 min\n\n## Ingredients\n- 200 g spaghetti or linguine\n- 80 ml extra virgin olive oil\n- 4 garlic cloves, thinly sliced\n- 1–2 small dried red chilies, crushed (or ¼ tsp chili flakes)\n- Salt, for pasta water\n- 15 g fresh flat-leaf parsley, chopped\n- Grated Parmesan or Pecorino (optional, traditionalists skip it)\n\n## Method\n1. **Boil pasta:** Bring a large pot of well-salted water to a rolling boil. Cook the pasta until al dente (about 1 minute less than package instructions). Reserve 1 cup of pasta water before draining.\n2. **Infuse oil:** While the pasta cooks, gently heat the olive oil in a wide pan over low-medium heat. Add the garlic and chilies, and cook slowly for 2–3 minutes until the garlic turns pale golden. **Don't let it brown** — it will turn bitter.\n3. **Combine:** Add the drained pasta to the pan with a splash of reserved pasta water. Toss vigorously over medium heat for about 1 minute so the pasta absorbs the oil.\n4. **Finish:** Add more pasta water as needed to create a light, silky sauce that coats the noodles. Stir in the parsley and season with salt.\n5. **Serve** immediately, topped with cheese if using.\n\n## Tips\n- Cold olive oil + slow heat = better flavor extraction.\n- The starch in the pasta water is key to a glossy, emulsified sauce.\n\nBuon appetito! 🍝",
        "name": "MiniMax AI",
        "audio_content": "",
        "reasoning_content": "The user is asking for a simple pasta aglio e olio recipe. The system prompt says I should give concise recipes with metric measurements. Let me provide a classic, simple recipe.",
        "reasoning_details": [
          {
            "type": "reasoning.text",
            "id": "reasoning-text-1",
            "format": "MiniMax-response-v1",
            "index": 0,
            "text": "The user is asking for a simple pasta aglio e olio recipe. The system prompt says I should give concise recipes with metric measurements. Let me provide a classic, simple recipe."
          }
        ]
      },
      "finish_reason": "stop"
    }
  ],
  "created": 1780336043,
  "model": "MiniMax-M3",
  "object": "chat.completion",
  "usage": {
    "total_tokens": 595,
    "prompt_tokens": 197,
    "completion_tokens": 398,
    "total_characters": 0,
    "prompt_tokens_details": {
      "cached_tokens": 114
    }
  },
  "input_sensitive": false,
  "output_sensitive": false,
  "base_resp": {
    "status_code": 0,
    "status_msg": ""
  },
  "input_sensitive_type": 0,
  "output_sensitive_type": 0,
  "output_sensitive_int": 0,
  "gatewayMetadata": {
    "keySource": "Unified"
  }
}

Multi-turn Conversation — Continuing a conversation with context

const response = await env.AI.run(
  'minimax/m3',
  {
    max_tokens: 2048,
    messages: [
      { content: 'What are the main differences between TCP and UDP?', role: 'user' },
      {
        content:
          'TCP is connection-oriented and guarantees delivery, while UDP is connectionless and faster but without delivery guarantees.',
        role: 'assistant',
      },
      { content: 'When would I choose UDP over TCP?', role: 'user' },
    ],
    temperature: 0.5,
  },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/v1/chat/completions \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "minimax/m3",
  "max_tokens": 2048,
  "messages": [
    {
      "content": "What are the main differences between TCP and UDP?",
      "role": "user"
    },
    {
      "content": "TCP is connection-oriented and guarantees delivery, while UDP is connectionless and faster but without delivery guarantees.",
      "role": "assistant"
    },
    {
      "content": "When would I choose UDP over TCP?",
      "role": "user"
    }
  ],
  "temperature": 0.5
}'

Use UDP for time-sensitive applications like video streaming, online gaming, or VoIP where speed matters more than perfect reliability.

{
  "id": "066cf8bb1c666fc68c6810518100c7ba",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Use UDP for time-sensitive applications like video streaming, online gaming, or VoIP where speed matters more than perfect reliability.",
        "name": "MiniMax AI",
        "audio_content": "",
        "reasoning_content": ""
      },
      "finish_reason": "stop"
    }
  ],
  "created": 1780336059,
  "model": "MiniMax-M3",
  "object": "chat.completion",
  "usage": {
    "total_tokens": 250,
    "prompt_tokens": 225,
    "completion_tokens": 25,
    "total_characters": 0,
    "prompt_tokens_details": {
      "cached_tokens": 114
    }
  },
  "input_sensitive": false,
  "output_sensitive": false,
  "base_resp": {
    "status_code": 0,
    "status_msg": ""
  },
  "input_sensitive_type": 0,
  "output_sensitive_type": 0,
  "output_sensitive_int": 0,
  "gatewayMetadata": {
    "keySource": "Unified"
  }
}

Creative Writing — Higher temperature for creative output

const response = await env.AI.run(
  'minimax/m3',
  {
    max_tokens: 2048,
    messages: [{ content: 'Write a haiku about programming.', role: 'user' }],
    temperature: 0.9,
  },
)
console.log(response)

curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/v1/chat/completions \
  --header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
  --header "Content-Type: application/json" \
  --data '{
  "model": "minimax/m3",
  "max_tokens": 2048,
  "messages": [
    {
      "content": "Write a haiku about programming.",
      "role": "user"
    }
  ],
  "temperature": 0.9
}'

Fingers on the keys,
Building worlds from empty void—
Code becomes the light.

{
  "id": "066cf8be5675980f2ff647974879fb77",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Fingers on the keys,\nBuilding worlds from empty void—\nCode becomes the light.",
        "name": "MiniMax AI",
        "audio_content": "",
        "reasoning_content": "The user wants a haiku about programming. A haiku is a traditional Japanese poem with three lines following a 5-7-5 syllable pattern. I should create one that captures something meaningful about programming.\n\nLet me think of a few options:\n\n1. About debugging:\nLines (5-7-5):\n- \"Bugs hide in the code\" (4) - need 5\n- \"Silent errors whisper\" (6) - need 7\n- \"Find them, fix, deploy\" (5) ✓\n\nLet me recount:\n- \"Bugs hide in the code\" - Bugs(1) hide(2) in(3) the(4) code(5) = 5 ✓\n- \"Silent errors whisper\" - Si-lent(2) er-rors(4) whis-per(6) = 6, need 7\n- Try: \"Silent errors always whisper\" - 7 ✓\n- \"Find them, fix, deploy\" - Find(1) them(2) fix(3) de-ploy(5) = 5 ✓\n\n2. About code in general:\n- \"Logic flows like streams\" (5) - Lo-gic(2) flows(3) like(4) streams(5) = 5 ✓\n- \"Through brackets and curved braces\" (7) - Through(1) brack-ets(3) and(4) curved(5) bra-ces(7) = 7 ✓\n- \"Dreams made into life\" (5) - Dreams(1) made(2) in-to(4) life(5) = 5 ✓\n\n3. Another option:\n- \"Fingers on the keys\" (5) - Fin-gers(2) on(3) the(4) keys(5) = 5 ✓\n- \"Building worlds from nothing\" (7) - Build-ing(2) worlds(3) from(4) no-thing(6) = 6, need 7\n- Try: \"Building worlds from empty void\" - Build-ing(2) worlds(3) from(4) emp-ty(6) void(7) = 7 ✓\n- \"Code becomes the light\" (5) - Code(1) be-comes(3) the(4) light(5) = 5 ✓\n\nLet me go with option 2 or 3, both are nice. I'll provide one clean haiku and maybe offer an alternative.\n\nActually, let me just give one good one. Option 3 feels more poetic and evocative.\n\nLet me verify option 3 once more:\n- \"Fingers on the keys\" - Fin-gers(2) on(3) the(4) keys(5) = 5 ✓\n- \"Building worlds from empty void\" - Build-ing(2) worlds(3) from(4) emp-ty(6) void(7) = 7 ✓\n- \"Code becomes the light\" - Code(1) be-comes(3) the(4) light(5) = 5 ✓\n\nThat works well. I could also provide a simpler one focused on a different aspect of programming.\n\nI'll give a haiku with a clean delivery. Let me also think about a second option to make it more useful, or just stick with one well-crafted one.\n\nI'll provide one haiku and keep it concise.",
        "reasoning_details": [
          {
            "type": "reasoning.text",
            "id": "reasoning-text-1",
            "format": "MiniMax-response-v1",
            "index": 0,
            "text": "The user wants a haiku about programming. A haiku is a traditional Japanese poem with three lines following a 5-7-5 syllable pattern. I should create one that captures something meaningful about programming.\n\nLet me think of a few options:\n\n1. About debugging:\nLines (5-7-5):\n- \"Bugs hide in the code\" (4) - need 5\n- \"Silent errors whisper\" (6) - need 7\n- \"Find them, fix, deploy\" (5) ✓\n\nLet me recount:\n- \"Bugs hide in the code\" - Bugs(1) hide(2) in(3) the(4) code(5) = 5 ✓\n- \"Silent errors whisper\" - Si-lent(2) er-rors(4) whis-per(6) = 6, need 7\n- Try: \"Silent errors always whisper\" - 7 ✓\n- \"Find them, fix, deploy\" - Find(1) them(2) fix(3) de-ploy(5) = 5 ✓\n\n2. About code in general:\n- \"Logic flows like streams\" (5) - Lo-gic(2) flows(3) like(4) streams(5) = 5 ✓\n- \"Through brackets and curved braces\" (7) - Through(1) brack-ets(3) and(4) curved(5) bra-ces(7) = 7 ✓\n- \"Dreams made into life\" (5) - Dreams(1) made(2) in-to(4) life(5) = 5 ✓\n\n3. Another option:\n- \"Fingers on the keys\" (5) - Fin-gers(2) on(3) the(4) keys(5) = 5 ✓\n- \"Building worlds from nothing\" (7) - Build-ing(2) worlds(3) from(4) no-thing(6) = 6, need 7\n- Try: \"Building worlds from empty void\" - Build-ing(2) worlds(3) from(4) emp-ty(6) void(7) = 7 ✓\n- \"Code becomes the light\" (5) - Code(1) be-comes(3) the(4) light(5) = 5 ✓\n\nLet me go with option 2 or 3, both are nice. I'll provide one clean haiku and maybe offer an alternative.\n\nActually, let me just give one good one. Option 3 feels more poetic and evocative.\n\nLet me verify option 3 once more:\n- \"Fingers on the keys\" - Fin-gers(2) on(3) the(4) keys(5) = 5 ✓\n- \"Building worlds from empty void\" - Build-ing(2) worlds(3) from(4) emp-ty(6) void(7) = 7 ✓\n- \"Code becomes the light\" - Code(1) be-comes(3) the(4) light(5) = 5 ✓\n\nThat works well. I could also provide a simpler one focused on a different aspect of programming.\n\nI'll give a haiku with a clean delivery. Let me also think about a second option to make it more useful, or just stick with one well-crafted one.\n\nI'll provide one haiku and keep it concise."
          }
        ]
      },
      "finish_reason": "stop"
    }
  ],
  "created": 1780336062,
  "model": "MiniMax-M3",
  "object": "chat.completion",
  "usage": {
    "total_tokens": 912,
    "prompt_tokens": 183,
    "completion_tokens": 729,
    "total_characters": 0,
    "prompt_tokens_details": {
      "cached_tokens": 114
    }
  },
  "input_sensitive": false,
  "output_sensitive": false,
  "base_resp": {
    "status_code": 0,
    "status_msg": ""
  },
  "input_sensitive_type": 0,
  "output_sensitive_type": 0,
  "output_sensitive_int": 0,
  "gatewayMetadata": {
    "keySource": "Unified"
  }
}

Parameters

▶messages[]

arrayrequired

temperature

numberminimum: 0maximum: 1

max_tokens

numberexclusiveMinimum: 0

max_completion_tokens

numberexclusiveMinimum: 0

top_p

numberminimum: 0maximum: 1

stream

boolean

▶stream_options{}

object

▶tools[]

array

tool_choice

stringenum: none, auto

▶response_format{}

object

mask_sensitive_info

boolean

string

▶choices[]

array

created

number

model

string

object

stringenum: chat.completion, chat.completion.chunk

▶usage{}

object

input_sensitive

boolean

output_sensitive

boolean

▶base_resp{}

object

API Schemas (Raw)

Input

Output