MiniMax M3.1 Flash: Behind The Open-Weight Leaderboard, Ahead In Our Newsroom Test

🔍 Read the full analysis: MiniMax M3.1 Flash: Behind The Open-Weight Leaderboard, Ahead In Our Newsroom Test on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get movie nights 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, positioning it as a fast, low-cost model for agent work at $0.15 per million input tokens and $0.60 per million output tokens. The available review says independent testing placed it behind several open-weight competitors, while the review’s account of a 24-article newsroom test does not provide the model’s scores.

The model is available through the MiniMax API, M Plan subscription, MiniMax Code and OpenRouter. OpenRouter offered it free for its first 72 hours. MiniMax says the price per token is half that of its predecessor, M3, and that the cost per agent task is about 70% lower. The latter figure is a company claim; the supplied material does not provide a methodology for independently checking it.

MiniMax advertises a one-million-token context window and adjustable reasoning effort. It says everyday output speed exceeds 200 tokens per second, with a median above 170, and describes the model as “nearly three times faster” than M3. Artificial Analysis measured 120 tokens per second and 1.08 seconds to first token in its evaluation, lower than the company’s speed claim. The difference may reflect test conditions; the supplied information does not establish a direct, like-for-like comparison.

The model’s weights have not yet been released. MiniMax says they will arrive on October 23, but has not stated a license or official parameter count. Artificial Analysis lists M3.1 Flash as proprietary while the weights remain unavailable. A reported estimate of 428 billion total parameters and 23 billion active parameters is not confirmed by MiniMax and should not be treated as established.

Lower Prices, Mixed Benchmark Standing

M3.1 Flash brings a lower published token price to a field where developers compare not just price, but also speed, reliability and the amount of text a model needs to complete a task. Artificial Analysis measured the model at 43 on its Intelligence Index, below MiMo-V2.6-Pro at 46, DeepSeek V4.1 Flash 1017 and GLM-5.3 at 45, and Step 5 Preview and Kimi K3 at 44.

Efficiency complicates the price comparison. Artificial Analysis says M3.1 Flash used 180 million tokens to complete its index, compared with a median of 99 million across the models it tested. That is a raw count in the cited evaluation, not a general estimate of token use for every task. It does suggest that developers should compare total task cost and output quality, rather than relying on per-token rates alone.

For teams using models to draft news, the practical question is whether low cost comes with faithful handling of sources. The supplied review describes a newsroom test but ends before reporting its results. Readers therefore cannot use this material to judge M3.1 Flash’s accuracy or compare it with other models on that test.

From Space Bunny to M3.1 Flash

Before MiniMax named the model, it ran anonymously on OpenRouter as Space Bunny Alpha. The provider listing appeared on September 23 and led OpenRouter’s usage ranking for 10 days before being removed on October 5. The review says it processed 56.7 trillion tokens on OpenRouter during that period; MiniMax separately counts more than 200 trillion tokens overall, without the supplied material defining the scope or period for that total.

MiniMax’s published benchmark charts show a mixed comparison with selected competitors. M3.1 Flash scored 62.2 on Automation-Bench v1.0.6, ahead of the listed rivals, while DeepSeek V4.1 Flash 1017 scored higher on DeepSWE and Terminal-Bench. On the listed HLE no-tools test, M3.1 Flash scored 43.4, below GPT-5.6 Sol and Claude Opus 5. These figures come from MiniMax’s own benchmark presentation; the review says the company chose the rivals and that the results have not been independently reproduced.

The review also flags a qualification around MiniMax’s claim that M3.1 Flash ranks first among Chinese models on Agents’ Last Exam. MiniMax’s chart does not show DeepSeek in that row, while DeepSeek’s model card reports a score of 35.4 for its 1017 model, potentially under a different setup. The comparison cannot be settled from the supplied figures alone.

““nearly three times faster” than M3”

— MiniMax

Weights, Data and Test Results

Several points remain open. MiniMax has not published the weights, their license or an official parameter count. Its October 23 release date is a promise, not a completed release in the material provided. The report does not establish whether that date was met.

The source says that, during the anonymous OpenRouter period, the provider could retain prompts but was not allowed to use them for training. MiniMax has not said what happened to that data. The available information does not state how long prompts were retained, whether they were deleted, or what access controls applied.

The newsroom benchmark setup is described as 12 real news sources, each turned into an English and German article, for 24 outputs. Two judges—GPT-6.1 Sol and Claude Opus 5.5—were to score source fidelity, invented facts, clarity, usefulness, structure and polish on a five-point scale. But the supplied source cuts off at the results table heading and includes no scores or conclusions. No newsroom performance finding can be reported from it.

Watch for Weights and Full Results

The next stated milestone is MiniMax’s promised October 23 release of the weights. Their publication, license and technical documentation would clarify whether developers can run or adapt the model outside MiniMax’s hosted services. The source does not confirm that the release has occurred.

For a fuller assessment, readers would also need the missing newsroom results, including the language-by-language scores and any comparison with other models under the same prompts. Independent replication of MiniMax’s benchmark claims and clearer information about the handling of prompts from the Space Bunny period would help answer questions the current account leaves unresolved.

Until those details are available, the confirmed picture is limited: M3.1 Flash is released at a low stated token price, its independent index score trails several listed open-weight rivals, and its advertised speed exceeds the speed measured by Artificial Analysis. How it performs on the newsroom test cannot be determined from the supplied results.

Key Questions

When was MiniMax M3.1 Flash released?

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?

The listed price is $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. OpenRouter offered the model free for its first 72 hours.

Are the model weights available?

Not in the supplied report. MiniMax said the weights would arrive on October 23, but the report does not confirm whether they were released. The company had not announced a license or official parameter count.

How did M3.1 Flash perform on independent testing?

Artificial Analysis gave it a 43 Intelligence Index score, below several models listed in the same comparison. It measured output speed at 120 tokens per second and 1.08 seconds to first token. Those results differ from MiniMax’s higher speed claims, though the source does not establish identical test conditions.

What did the newsroom test find?

The supplied material describes a 24-article test judged by GPT-6.1 Sol and Claude Opus 5.5, but omits the results table and scores. Its findings cannot be confirmed from the information available here.

Source: ThorstenMeyerAI.com