MiniMax M3.1-Flash text generation and model configuration

One-sentence takeaway

MiniMax-M3.1-Flash targets agents, tool calling, coding, and long-context tasks. Text generation is available through Anthropic-compatible or OpenAI-compatible APIs, with effort / reasoning_effort controlling adaptive thinking depth.

Use cases

• Suitable tasks: Agent workflows, tool calling, coding, complex tasks, long-context text generation, and multimodal chat input.

• Unsuitable tasks: Production configurations that need thinking disabled or depend on undocumented limits, pricing, maximum output limits, or other parameters.

• Applicable model version: MiniMax-M3.1-Flash. The page also lists MiniMax-M3, but this configuration is not extrapolated to other models.

• Applicable clients, agents, or APIs: MiniMax API, Anthropic SDK, OpenAI SDK, and agents or tools such as Claude Code and Cursor that support the relevant protocols.

• Recommended reasoning tiers and parameters: effort / reasoning_effort can be low, medium, high, xhigh, or max. The page says omitted values default to max; the examples explicitly use max.

Ready-to-use content

Model/API ID: MiniMax-M3.1-Flash

Basic configuration

The page lists a context window of 1,000,000 tokens. It describes the model as a multimodal coding model with adjustable thinking depth.

Anthropic-compatible request

The official documentation marks this path as recommended. It supports thinking blocks, Interleaved Thinking, and streaming output. The official fields and values are preserved below. Set stream to true to switch to a streaming response.

curl https://api.minimax.io/anthropic/v1/messages \
  -H "Authorization: Bearer <MINIMAX_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-M3.1-Flash",
    "output_config": {"effort": "max"},
    "max_tokens": 4096,
    "messages": [
      {"role": "user", "content": "Hi, how are you?"}
    ]
  }'

OpenAI-compatible request

Clients that already use the OpenAI SDK can replace base_url and model. The page says that this protocol always returns thinking content in reasoning_content and the final answer in content, so there is no need to parse <think> tags.

curl https://api.minimax.io/v1/chat/completions \
  -H "Authorization: Bearer <MINIMAX_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMax-M3.1-Flash",
    "reasoning_effort": "max",
    "messages": [
      {"role": "user", "content": "Hi, how are you?"}
    ]
  }'

Thinking-depth field mapping

Minimal explicit configuration for Anthropic-compatible:

{
  "model": "MiniMax-M3.1-Flash",
  "max_tokens": 4096,
  "output_config": {"effort": "max"},
  "messages": [{"role": "user", "content": "Hello"}]
}

Test or workflow steps

1. Obtain a MiniMax Subscription Key and inject it into the runtime environment as MINIMAX_API_KEY.

2. Choose a protocol: prefer https://api.minimax.io/anthropic; use https://api.minimax.io/v1 with an existing OpenAI SDK.

3. Set model to MiniMax-M3.1-Flash. Do not substitute the MiniMax-M3 listed on the page.

4. Set effort or reasoning_effort based on latency, output length, and task complexity. The page says the default is max when the field is omitted.

5. For streaming output, set stream to true in the request and read the thinking content and final answer according to the selected protocol.

6. Send a short text request to confirm authentication, the model ID, protocol fields, and response fields before connecting an agent or tool-calling workflow.

Raw evidence and data

• In its Supported Models table, the page lists MiniMax-M3.1-Flash with a 1,000,000-token context window, describes it as a multimodal coding model, and notes support for adjustable thinking depth.

• The page lists these model-specific points: 1M-token context, Agent and coding workflows, Tunable thinking depth (effort), and Multimodal chat input.

• The Anthropic-compatible path is https://api.minimax.io/anthropic/v1/messages; the OpenAI-compatible path is https://api.minimax.io/v1/chat/completions.

• The page says thinking is enabled by default. effort supports low, medium, high, xhigh, and max, and the page says the default is max when the field is omitted.

• The limitation error shown on the page is: model "MiniMax-M3.1-Flash" requires adaptive thinking.

• The page does not publish a release date, pricing, rate limits, maximum output tokens, complete SDK code, or a unified upper limit for tool-calling parameters. These values cannot be inferred from this page.

Scope and limitations

• Thinking mode cannot be disabled. Sending thinking: {"type": "disabled"} or effort: "none" returns HTTP 400. Lower effort to reduce latency and thinking tokens.

• max_tokens: 4096 is the value in the official example. It does not represent the model's maximum output limit.

• The page marks Anthropic-compatible as recommended but does not describe all feature differences between the two protocols. Test against the actual SDK and response structure before integration.

• Although the page mentions multimodal chat input, it does not provide a message structure for images or other modalities. Do not add a request format based on this mention alone.

• This material records API configuration. It does not mean MiniMax-M3.1-Flash is available in Tabbit's model selector.

Reproduction notes

On 2026-10-21, the original page was opened and read through Tabbit's dedicated task M31 prompt01 text. The page was accessible, and its main text was extracted to /tmp/m31-text-evidence.txt. The relevant official navigation on the page includes:

• Documentation index: https://platform.minimax.io/docs/llms.txt

• Model invocation: https://platform.minimax.io/docs/guides/text-generation

• Tool calling and interleaved thinking: https://platform.minimax.io/docs/guides/text-m3-function-call

• Prompt Caching: https://platform.minimax.io/docs/api-reference/text-prompt-caching

• Anthropic API Compatible: https://platform.minimax.io/docs/api-reference/text-anthropic-api

• OpenAI API Compatible: https://platform.minimax.io/docs/api-reference/text-openai-api

• SDK quick start: https://platform.minimax.io/docs/guides/quickstart-sdk

• M-series models and coding tools: https://platform.minimax.io/docs/m-plan/openclaw

• API Tokens Per Second explanation: https://platform.minimax.io/docs/faq/about-apis#q-how-is-tps-tokens-per-second-calculated-for-llms

• MiniMax M3.1-Flash product page: https://www.minimax.io/blog/minimax-m31f