MiniMax officially rolled out M3.1 Flash on October 19, targeting coding, computer use, and multi-step agent workflows. It is available through the MiniMax API, M Plan, and MiniMax Code. Downloadable weights are promised later—not part of the current release.
The practical pricing details
MiniMax lists the following API rates in US dollars per million tokens:
- Input: $0.15.
- Output: $0.60.
- Cached input reads: $0.015.
These are token prices, not the price of completing a task. Actual usage depends on the conversation, generated reasoning, tool loop, and retries.
Long context with deliberate reasoning control
M3.1 Flash offers a one-million-token context window and adjustable reasoning depth. In the Anthropic-compatible API, output_config.effort accepts low, medium, high, xhigh, and max.
The default is max when that setting is omitted. Developers should choose the effort level deliberately: higher settings produce more thinking tokens and can increase latency. Thinking cannot be disabled for this model through that interface.
Vendor-reported gains, not guaranteed savings
MiniMax says online inference is nearly three times faster and overall agent-task usage costs are 70% lower than its predecessor. These are company claims, not independent confirmation or guaranteed savings for every workflow.
Why it matters for builders
Agents make repeated model calls while reading files, using tools, and checking results. Lower token prices could make those loops more affordable—but retries and reasoning effort can erase the advantage.
That cost-per-task perspective also matters when evaluating updates such as DeepSeek’s V4.1 Flash coding-agent release: headline pricing and completed work are different measures.
For builders such as SunMarc App Labs, the useful next test is a bounded app-maintenance or website task: compare completion quality, elapsed time, and total cost per accepted result before switching models. This is a proposed evaluation, not an announced SunMarc deployment or a claim of tested performance.