MiniMax M3.1 Flash: Is It Free? API Pricing, Benchmarks and How to Use It
MiniMax M3.1 Flash launched on October 19, 2026, and you can use it today through the MiniMax API, M Plan or MiniMax Code. The API costs $0.15 per 1M input tokens and $0.60 per 1M output tokens. There is no confirmed free tier, the model weights are not downloadable yet, and aiwith does not host M3.1 Flash.
Last checked: October 20, 2026. This is an access and pricing guide based on MiniMax's official release post, not a hands-on benchmark review.
Is MiniMax M3.1 Flash free?
No free route has been confirmed. MiniMax lists three ways to use the model right now, and all of them are commercial products:
- MiniMax API at platform.minimax.io, billed per token.
- M Plan, MiniMax's subscription plan, at minimax.io/m-plan.
- MiniMax Code, the desktop agent app, downloadable from agent.minimax.io/download.
If a free trial or promotional credit appears in your MiniMax account, check it inside the console before relying on it. Nothing in the official launch post promises free chat access or free API calls.
The open-weight question is separate. MiniMax says model weights will be released later but does not give a date in the launch post. Some community trackers have repeated October 23 as a possible weights date; treat that as unconfirmed until MiniMax publishes the files. Free weights would still mean paying for GPUs to run them, not free hosted inference.
What is MiniMax M3.1 Flash?
M3.1 Flash is MiniMax's new speed-and-cost model, positioned below its Pro tier. MiniMax says an early anonymous version, codenamed Space Bunny, processed more than 200 trillion tokens before launch.
The headline claims in the official post are:
- Nearly 3x faster online inference than the previous generation.
- 70% lower usage cost for agentic tasks.
- 200+ output tokens per second, with a median (P50) of 170+ TPS.
- Improvements in agentic work, coding, computer use and cybersecurity.
Those are vendor claims. We have not independently benchmarked the model, and speed will vary by provider, region and prompt shape.
Benchmarks: what MiniMax reported
MiniMax published these scores in its launch post:
| Benchmark | What it measures | M3.1 Flash |
|---|---|---|
| BankerToolBench | Multi-step planning, document consistency, financial compliance | 76.2 |
| DeepSWE v1.1 | Real-world software engineering with independent testing | 73.7% |
| OSWorld V2 | Computer use and cross-application tasks | 61.7 |
| Agents' Last Exam (ALE) | Real professional work with economic value | 29.6 |
| CyberGym | Vulnerability discovery and validation | 82% |
MiniMax also says M3.1 Flash ranks among the top performers on AutomationBench-AA, which tests end-to-end enterprise workflows across SaaS systems, but it did not publish a headline score for that benchmark in the post. The company claims first place among Chinese models on OSWorld V2 and ALE. All of these numbers are self-reported, and benchmark definitions change, so compare them with independent leaderboards before making a switching decision.
Pricing: what M3.1 Flash costs
MiniMax lists flat per-1M-token pricing:
| Model | Input | Output | Cached input |
|---|---|---|---|
| MiniMax-M3.1-Flash | $0.15 | $0.60 | $0.015 |
That is the direct MiniMax API price. A simple mix of 1M input tokens and 1M output tokens costs about $0.75 before any cache savings. Agent workloads tend to be input-heavy because tool results and long contexts are resent, so the $0.015 cached-input rate matters more than the headline output price.
M Plan bundles access into a subscription instead of per-token billing. Compare the plan's included usage with your actual token volume before assuming it is cheaper than pay-as-you-go.
How do I use MiniMax M3.1 Flash?
Pick the route that matches what you need:
- Direct API: create a key at platform.minimax.io, then call the model listed in MiniMax's pricing table as
MiniMax-M3.1-Flash. Confirm the exact model ID in the current API docs before wiring it into production. - M Plan: subscribe at minimax.io/m-plan if you want usage bundled into a plan rather than metered billing.
- MiniMax Code: download the desktop app from agent.minimax.io/download if you want MiniMax's own agent interface rather than an API integration.
- Self-hosting: wait. The weights are not available yet, and MiniMax has not published a license or hardware guide for them.
If you are building an agent, budget for tool-call overhead. A model that is cheap per token can still produce a large bill when it retries tools or carries long transcripts between steps.
Can I use M3.1 Flash on aiwith or Router?
Not yet on aiwith. aiwith's model line-up does not include MiniMax M3.1 Flash, so an aiwith subscription does not unlock it. If you want a comparable low-cost model inside aiwith today, these pages map to similar jobs:
- DeepSeek V4.1 — the value option for coding, agent workflows and everyday chat.
- Kimi K3 — long-context work and document-heavy tasks.
- GPT and Claude — frontier proprietary models when the cheap tier is not enough.
On Router, M3.1 Flash is not listed yet. Router does carry MiniMax models today, including minimax/MiniMax-M3 with a 1M-token context window and tool calling. Router's listed price for M3 is $4.20 per 1M input tokens and $16.80 per 1M output tokens, with cached input at $0.84. That is MiniMax-M3, not M3.1 Flash — a different model with different pricing. You can call it through Router's OpenAI-compatible endpoint:
curl https://router.bumo.cc/api/v1/chat/completions \
-H "Authorization: Bearer $ROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax/MiniMax-M3",
"messages": [{"role": "user", "content": "Summarize this changelog."}]
}'
See Router for the current model list and key setup. If MiniMax M3.1 Flash is added later, the model list is the authoritative place to check.
Should you switch to M3.1 Flash?
M3.1 Flash is aimed at high-volume agent and coding work where cost per completed task matters more than raw benchmark leadership. At $0.15/$0.60, its token prices are far below most frontier models, and the 170+ TPS median suggests it can keep an agent loop responsive.
Three caveats before you migrate:
- The benchmark numbers are vendor-reported. Wait for independent evaluations on your own workload.
- Weights are not out. If self-hosting or license terms matter to you, this is not yet a decision you can make.
- Hosted access is not free. Budget from the API price, not from the expectation of a free tier.
If you need a MiniMax model through an OpenAI-compatible gateway today, Router's MiniMax-M3 is available now. If you specifically need M3.1 Flash, use MiniMax's own API until it appears in the gateways you rely on.