Why We Tested MiniMax M3.1 Flash Beyond The Leaderboard

🔍 Read the full analysis: Why We Tested MiniMax M3.1 Flash Beyond The Leaderboard on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

MiniMax released M3.1 Flash on October 19, presenting it as a low-cost, fast model for agent tasks, while a review compared the company’s claims with Artificial Analysis measurements and a 24-article newsroom test. The available results point to a model with competitive pricing and solid benchmark scores, but do not establish that it leads its open-weight rivals on speed, coding or real-world news writing.

MiniMax lists M3.1 Flash at $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. That is half the listed per-token price of its predecessor, M3, according to the review. MiniMax says the model costs about 70% less per agent task, but that figure is a company claim rather than an independently verified result. The model is available through MiniMax’s API, M Plan subscription, MiniMax Code and OpenRouter; OpenRouter offered free access for its first 72 hours.

The company claims everyday generation speeds above 200 tokens per second, with a median above 170, and says the model is nearly three times faster than M3. Artificial Analysis measured 120 tokens per second and 1.08 seconds to first token. Its Intelligence Index score was 43, below scores of 44 to 46 for several listed open-weight competitors. Artificial Analysis also found that M3.1 Flash used 180 million tokens to complete its index evaluations, compared with a 99 million-token median.

The review also describes a newsroom evaluation using 24 articles in English and German, based on 12 real news sources and a production prompt. Two judges, GPT-6.1 Sol and Claude Opus 5.5, scored source fidelity, invented facts, clarity, usefulness, structure and polish. The supplied source material describes the method but omits the score table, so it does not support a conclusion about how M3.1 Flash performed in that test.

Price Meets Real-World Performance

M3.1 Flash’s low listed price could matter to developers running high-volume agent workflows, where token costs add up. But price per token alone does not determine the cost of a completed task: a model that uses more tokens may narrow the savings. Artificial Analysis’s finding that M3.1 Flash used 180 million tokens, against a 99 million median in its index, illustrates why buyers may need to compare total usage alongside unit prices.

The gap between MiniMax’s speed claim and the independent measurement also matters for teams weighing latency against cost. The review’s newsroom test addresses a separate concern: whether the model can turn source material into accurate, useful copy without adding unsupported details. Since the test’s scores are not included in the supplied material, that practical question remains open rather than settled.

From Space Bunny to M3.1

Before the public release, the model appeared anonymously on OpenRouter as Space Bunny Alpha on September 23. The review says it led OpenRouter’s usage ranking for 10 days, processed 56.7 trillion tokens on the platform and was removed on October 5. MiniMax says the model handled more than 200 trillion tokens overall. Those usage figures describe activity, not independently verified quality or accuracy.

MiniMax revealed the model’s identity on October 19 and said it was an early version of M3.1 Flash. The company has described its design only as using “more aggressive sparsity”. It says the model supports a one-million-token context window and adjustable reasoning effort. Its weights are scheduled for October 23; until they are released, Artificial Analysis lists the model as proprietary.

MiniMax’s published benchmark chart compares M3.1 Flash with selected competitors, including DeepSeek V4.1 Flash 1017, which the review says launched two days earlier at the same token price. M3.1 Flash scores higher on Automation-Bench, while DeepSeek leads on DeepSWE and Terminal-Bench. The review cautions that these chart results are self-reported, that MiniMax chose the comparison models, and that some listed rivals are not the newest available versions.

““nearly three times faster” than M3”

— MiniMax

Scores, Weights and Data Questions

The supplied review material does not include the newsroom benchmark’s results table. Although it says two independent judges scored 24 articles, the scores and model comparisons are missing, so the test’s outcome cannot be reported from this source. MiniMax’s benchmark charts have not been independently reproduced in the material, and its security results, including a claimed 82% score on CyberGym, are described as internal evaluations.

The model’s parameter count and license remain unconfirmed. One outlet reportedly gave a 428-billion total and 23-billion active parameter count, but MiniMax has not confirmed those figures. It is also unclear whether the promised weights will include a license permitting commercial use. The review raises a further data question: while Space Bunny Alpha was hosted anonymously, the provider could retain prompts but was not allowed to train on them. MiniMax has not said what happened to that data after the model was identified.

October 23 Weights Release

The next stated milestone is October 23, when MiniMax says it plans to publish M3.1 Flash’s weights. Their release, together with the licensing terms and any confirmed model specifications, will help clarify whether developers can run the model independently and what uses are permitted. Further independent testing could also establish how its speed, token consumption and task-level costs compare with competitors under consistent conditions.

For the newsroom evaluation, readers would need the missing score table and comparison-model results to judge whether M3.1 Flash met the review’s production criteria. Until those results are available, the evidence supports conclusions about the model’s announced price and the cited index measurements, but not about its relative performance on the article-writing test.

Key Questions

When did MiniMax release M3.1 Flash?

MiniMax released M3.1 Flash on October 19, 2026, through its API, M Plan subscription, MiniMax Code and OpenRouter.

How much does M3.1 Flash cost?

MiniMax lists the price at $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. Its claim that the model costs about 70% less per agent task has not been independently verified in the supplied material.

Is M3.1 Flash faster than its competitors?

MiniMax claims speeds above 200 tokens per second in everyday use. Artificial Analysis measured 120 tokens per second, below that claim and below its reported 211 tokens per second for DeepSeek V4.1 Flash 1017.

Did the newsroom test show that M3.1 Flash writes accurate articles?

The review describes a 24-article test judged for accuracy and other qualities, but the supplied material does not include the scores. Its performance on that test therefore cannot be established from the available information.

When will M3.1 Flash’s weights be available?

MiniMax said the weights would arrive on October 23, 2026. The supplied material does not include the eventual release or its license terms.

Source: ThorstenMeyerAI.com