MiniMax M3.1 Flash: Open-Weight Rankings Meet A Newsroom Test

🔍 Read the full analysis: MiniMax M3.1 Flash: Open-Weight Rankings Meet A Newsroom Test on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

MiniMax released M3.1 Flash on October 19, offering the model through its API, M Plan subscription, MiniMax Code and OpenRouter at $0.15 per million input tokens and $0.60 per million output tokens. Early comparisons point to a low-cost model that is not at the top of an independent open-weight ranking, while a separate newsroom evaluation tested its performance on 24 real news articles.

MiniMax says M3.1 Flash is designed for fast, low-cost agent work and costs half as much per token as its predecessor, M3. The company also claims it costs about 70% less per agent task. Those are MiniMax’s figures; the source report does not provide enough detail about the task comparisons to independently verify the claim. OpenRouter offered the model free for its first 72 hours.

The company advertises a one-million-token context window and says everyday output speed exceeds 200 tokens per second, with a median above 170. Artificial Analysis measured 120 tokens per second and 1.08 seconds to first token. That result is below MiniMax’s stated speed, though the measurements come from different testing setups. MiniMax has attributed the speed improvement to “more aggressive sparsity” without publishing further design detail.

On Artificial Analysis’s Intelligence Index, M3.1 Flash scored 43. The index listed DeepSeek V4.1 Flash 1017 and GLM-5.3 at 45, and MiMo-V2.6-Pro at 46. Artificial Analysis also measured 180 million tokens used to complete its index evaluation, compared with a 99 million median. That figure is a measure of the test’s token use, not a direct measure of quality or cost across every use case.

The report also describes a newsroom test using 12 source items in English and German, producing 24 articles. Two judges, GPT-6.1 Sol and Claude Opus 5.5, were to score each article on source fidelity, invented facts, clarity, usefulness, structure and polish. The supplied source material does not include the scores or findings, so no conclusion about M3.1 Flash’s newsroom performance can be drawn from it.

Price Versus Tested Performance

M3.1 Flash’s low listed token price may appeal to developers running repeated agent tasks, but price per token is only one part of the cost. Artificial Analysis found the model used 180 million tokens in its index evaluation, against a 99 million median. A model that consumes more tokens can narrow the savings implied by its headline rates, depending on the task and the user’s billing.

The results also show why separate evidence matters. MiniMax’s published comparisons are company-selected and self-reported; Artificial Analysis provides a third-party index; and the newsroom test targets factual reporting rather than general intelligence. None alone answers whether the model is the right choice for a particular workflow. The missing newsroom scores and unreleased weights leave important questions about practical quality and openness unanswered.

From Space Bunny to M3.1

Before MiniMax identified the model, it appeared on OpenRouter under the anonymous name Space Bunny Alpha. According to the source report, it went live on September 23, led OpenRouter’s usage ranking for 10 days, processed 56.7 trillion tokens there and was removed on October 5. MiniMax says it processed more than 200 trillion tokens overall; that broader figure is the company’s own account.

The report says the anonymous OpenRouter period allowed the provider to retain prompts, though not to use them for training. MiniMax has not clarified what happened to that data. The company says the model’s weights will arrive October 23, but at release it had not published a license or official parameter count. A parameter estimate of 428 billion total and 23 billion active has appeared in another outlet, but MiniMax has not confirmed it.

MiniMax’s own charts compare M3.1 Flash with selected competitors, including DeepSeek V4.1 Flash 1017. In those results, DeepSeek scored higher on DeepSWE and Terminal-Bench, while M3.1 Flash led Automation-Bench by 1.7 points. The report notes that some newer competing models were not included. The figures remain company-reported rather than independently reproduced.

““more aggressive sparsity””

— MiniMax

Weights, Data and Test Results

The model’s weights, license and official parameter count were not available at launch. Until the weights are released and their terms are published, it is not possible to assess how open the release will be or independently inspect the model. MiniMax’s claimed speed and agent-task savings also have not been established by the independent measurements described in the source.

There is also no disclosed answer about the prompts retained during the Space Bunny period. The source report does not say how long those prompts were kept, whether they were later deleted, or what users were told at the time. Finally, the newsroom benchmark setup is described, but its scores and comparative results are absent from the supplied material. Its outcome is therefore unknown.

October 23 Weights Release

The next stated milestone is October 23, when MiniMax says it will release the weights. The release details should clarify the model’s license and may provide the parameter information that was missing at launch. Until then, Artificial Analysis lists M3.1 Flash as proprietary because its weights are unavailable.

Further reporting can compare the released model with the published claims and assess whether independent testing reproduces its performance. The newsroom evaluation’s scores would also be needed before drawing conclusions about its ability to turn real sources into accurate articles. MiniMax has not provided a date for clarifying the handling of prompts from the anonymous OpenRouter period.

Key Questions

How much does MiniMax M3.1 Flash cost?

MiniMax lists $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. OpenRouter offered it free for its first 72 hours.

How did it perform on the independent index?

Artificial Analysis gave M3.1 Flash an Intelligence Index score of 43. The index listed several models above it, including DeepSeek V4.1 Flash 1017 and GLM-5.3 at 45.

Are the model weights available?

Not at the October 19 launch described in the source. MiniMax said the weights would arrive October 23; the release license and official parameter count were not yet disclosed.

What did the newsroom test find?

The test ran the model on 24 articles based on 12 real sources, with English and German versions scored by two judges. The supplied source material gives the methodology but not the scores, so its findings cannot be reported.

What is known about prompts from its Space Bunny period?

The report says the provider could retain prompts during the anonymous OpenRouter period but could not train on them. MiniMax has not said what happened to the retained data, including whether it was deleted.

Source: ThorstenMeyerAI.com