MiniMax M3.1 Flash vs DeepSeek: How to Try It Free Until Oct 22

MiniMax M3.1 Flash vs DeepSeek key facts: free on OpenRouter and Ollama until Oct 22, $0.15 input and $0.60 output per million tokens, 73.7% vs 77.0% on DeepSWE, open weights due Oct 23

Update, Oct 22, 2026: The free period is over on OpenRouter and OpenCode. OpenRouter says the free launch period ended on October 22 (12:00 UTC, 5 PM PKT) and the model is now billed at MiniMax's list price of $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens; the :free model ID has been retired, so switch to minimax/minimax-m3.1-flash. OpenCode's Zen model list no longer includes the free version. Ollama lists its free cloud access until 11:59 PM UTC on October 22 (4:59 AM PKT on October 23), after which the same rates apply. We found no extension announced by MiniMax.

MiniMax M3.1 Flash was free to try through OpenRouter, Ollama and OpenCode until October 22; it now costs $0.15 per million input tokens and $0.60 per million output tokens, the same as DeepSeek V4.1 Flash 1017's off-peak rate. In short, MiniMax wins on flat round-the-clock pricing, video input and business-automation scores in its own tests. DeepSeek still leads on the DeepSWE coding test and on an independent intelligence index, and it already ships open weights.

What happened

MiniMax released the model on Monday, October 19, in a launch post that also solved a small mystery: Space Bunny, the anonymous model OpenRouter says was its top model for 10 straight days, was an early version of M3.1 Flash. Reuters reported the launch the same day, noting MiniMax's claim that online inference is nearly three times faster and usage costs are 70% lower.

M3.1 Flash is a fast, low-cost model from the Chinese AI company MiniMax, built for agents, coding, computer use and security work. Below: how to use it now that the free period has ended, a side-by-side table against DeepSeek, and one catch to know before you trust it.

Advertisement

Key details

  • Context and inputs: A 1 million-token context window, with text, image and video input, according to MiniMax's API documentation.
  • Reasoning: Always on, with five effort levels; thinking cannot be turned off.
  • Size and speed: SiliconANGLE reports that MiniMax's preview developer guide lists 428 billion parameters. MiniMax says output can pass 200 tokens per second.
  • Open weights: MiniMax says the weights "will arrive on Oct 23," according to its launch post on X. Until then it is a hosted model only.

How to use MiniMax M3.1 Flash now (the free window has closed)

  1. Know that the free routes have closed. OpenRouter said at 12:01 UTC on Monday that the model was "free for the next 72 hours," and its free model page confirms the free launch period ended at 12:00 UTC on Thursday, October 22. Ollama's model page lists it as free until 11:59 PM UTC on October 22, OpenCode had said it was free in its app for 3 days, and the free version is no longer on its Zen model list.
  2. On OpenRouter, create an API key in your dashboard, add credits, point any OpenAI-compatible client at https://openrouter.ai/api/v1 and set the model to minimax/minimax-m3.1-flash (the retired :free ID was shown on the OpenRouter model page). The paid model page lists $0.15 input and $0.60 output per million tokens. For other free models, OpenRouter's free-model limits are 20 requests a minute and 50 a day (1,000 a day once you have bought at least 10 credits).
  3. On Ollama, run one command: ollama run minimax-m3.1-flash:cloud (free until 11:59 PM UTC on October 22, then billed at the listed rates). For Claude Code, use ollama launch claude --model minimax-m3.1-flash:cloud; Ollama lists similar commands for OpenCode and OpenClaw.
  4. Choose an effort level. MiniMax offers low, medium, high, xhigh and max, with max as the default. Go lower if answers are slow or burn too many tokens.
  5. Or go direct to the paid MiniMax API. Get a key at platform.minimax.io, set the base URL to https://api.minimax.io/v1 (OpenAI-style) or https://api.minimax.io/anthropic (Anthropic-style) and the model to MiniMax-M3.1-Flash. OpenRouter now charges the same $0.15 and $0.60 rates.
  6. Check the output. Test code and confirm facts before relying on it (see the hallucination results below).

OpenCode users should first read our OpenCode Console breach explainer to check whether saved provider keys may have been exposed.

Quick answers

Is MiniMax M3.1 Flash still free?

No. The free launch offers ended on October 22, 2026: OpenRouter retired the :free version at 12:00 UTC, OpenCode dropped it from its Zen list, and Ollama listed its free cloud access only until 11:59 PM UTC that day. If you need a free model, Nvidia's Nemotron 3.5 Super VL is still listed free on OpenCode Zen.

How much does MiniMax M3.1 Flash cost?

$0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens, at a flat rate.

Is MiniMax M3.1 Flash open source?

Not yet. MiniMax said on X that weights would arrive on October 23, while its launch blog says only that they "will be released later," and VentureBeat notes it has not shared licensing terms, so self-hosting rights are still unclear.

MiniMax M3.1 Flash vs DeepSeek V4.1 Flash 1017

Rows marked "company table" come from MiniMax's launch chart and are not independently checked. DeepSeek prices are from its pricing page.

FeatureMiniMax M3.1 FlashDeepSeek V4.1 Flash 1017
Input / output price per 1M tokens$0.15 / $0.60, flat$0.15 / $0.60 off-peak; $0.30 / $1.20 peak
Cached input per 1M tokens$0.015$0.003 off-peak; $0.006 peak
Context window1M tokens1M tokens
InputsText, image, videoText and images
ReasoningAlways on, 5 effort levelsThinking or non-thinking mode
Open weightsPromised for October 23; license not announcedAvailable now under the MIT license
DeepSWE v1.1 coding (company table)73.7%77.0%
Automation-Bench v1.0.6 (company table)62.260.5
Terminal Bench 4.0 (company table)38.142.3
Artificial Analysis Intelligence Index42, at about $0.22 per task45, at about $0.26 per task

DeepSeek's peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday; outside them, both cost the same per token. Our DeepSeek V4.1 Flash 1017 breakdown has more on that model.

Which one should you use?

  • Pick MiniMax M3.1 Flash if your agents run during DeepSeek's peak hours, you need video input, or your work is business automation, where MiniMax's own table puts it slightly ahead.
  • Pick DeepSeek V4.1 Flash 1017 if coding scores matter most, you need open weights today, or you reuse long prompts and want the cheaper cache.
  • Watch the hallucination risk. VentureBeat reports that Artificial Analysis found M3.1 Flash answered 56% of its knowledge questions correctly, versus 43% for DeepSeek, but showed a 93% hallucination rate. When it did not know, it nearly always gave a wrong answer instead of admitting uncertainty.
  • Government suppliers, take note. The US Department of Homeland Security has ordered federal civilian agencies to remove DeepSeek products, including third-party services built with DeepSeek, BleepingComputer reports. It is not a ban on private or consumer use; our DeepSeek federal ban guide explains who it covers.

Both are far cheaper than premium models: VentureBeat lists Claude Opus 5.5 at $4 input and $20 output per million tokens, and Claude Opus 5.5 Fast Mode costs even more.

What's next

  • October 22 (done): The free OpenRouter and OpenCode offers ended; Ollama's was listed to end at 11:59 PM UTC. No extension was announced.
  • October 23: The date MiniMax gave on X for the weights; its blog says "later." Watch for the release and the license terms.
  • Pro model: MiniMax says a more capable Pro version is coming later, with no date yet.
  • Independent tests: Wait for third-party coding results before moving production work.

FAQ

How do I use MiniMax M3.1 Flash for free?

Not any more. The launch offers on OpenRouter and OpenCode ended on October 22, 2026, and Ollama listed its free access only until 11:59 PM UTC that day. Use minimax/minimax-m3.1-flash on OpenRouter or the MiniMax API at $0.15 per million input tokens and $0.60 per million output tokens.

Is MiniMax M3.1 Flash better than DeepSeek V4.1 Flash?

It depends on the job. MiniMax leads on business-automation tests in its own results and offers flat pricing, while DeepSeek V4.1 Flash 1017 scores higher on DeepSWE coding and on the Artificial Analysis Intelligence Index.

When will MiniMax M3.1 Flash open weights be released?

MiniMax said on X that the weights would arrive on October 23, 2026; its launch blog says they "will be released later." It has not yet said which license they will use.

Does MiniMax M3.1 Flash support images and video?

Yes. It accepts text, image and video input and returns text. It does not generate images or video itself.

Related reading

Sources

Comments

Popular posts from this blog

Meta Muse Shared Chats Exposed: What to Do If You Shared One

OpenAI Agent Hacked an Australian Government Site: What We Know

Suncatcher TPU Satellite: How to Track It and What Happens Next