MiniMax M3.1 Flash: Our Newsroom Test Put Its Open-Model Performance To The Test

🔍 Read the full analysis: MiniMax M3.1 Flash: Our Newsroom Test Put Its Open-Model Performance To The Test 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, describing it as a lower-cost, faster model for agent work, but available evidence presents a mixed picture: the company’s claims are not yet independently reproduced, and Artificial Analysis measured 120 tokens per second, below MiniMax’s stated everyday speed of more than 200. The model is available through several services, while its weights are promised for October 23. A newsroom test using 24 real news articles was also conducted, but the source account does not provide the scores, so its findings cannot be assessed here.

MiniMax lists M3.1 Flash at $0.15 per million input tokens and $0.60 per million output tokens, with cached input priced at $0.015 per million. Those rates are half the listed per-token prices of its M3 predecessor, according to the source account. MiniMax also says the model costs about 70% less per agent task, a company claim whose test conditions are not provided in the material. The model is available through the MiniMax API, M Plan subscription, MiniMax Code and OpenRouter; OpenRouter offered it free for its first 72 hours.

The company says M3.1 Flash has a one-million-token context window, adjustable reasoning effort and everyday output speed above 200 tokens per second, with a median above 170. Artificial Analysis measured 120 tokens per second and 1.08 seconds to first token. Its Intelligence Index score was 43, compared with 45 for DeepSeek V4.1 Flash 1017 and GLM-5.3 in the cited comparison. Artificial Analysis currently classifies MiniMax’s model as proprietary because its weights have not been released.

The source account says its newsroom team ran the model on 12 real news sources in English and German, producing 24 articles with a production prompt. Two judges, GPT-6.1 Sol and Claude Opus 5.5, assessed source fidelity, invented facts, clarity, usefulness, structure and polish on a five-point scale. However, the supplied material ends before giving the results table. It does not establish how M3.1 Flash scored or how it compared with other models in that test.

Low Prices, Uneven Performance Evidence

M3.1 Flash’s headline appeal is its low token price paired with a long context window and a focus on agent tasks. For developers, those features could make experimentation or high-volume use less costly. But token price alone does not establish the cost of completing a task: Artificial Analysis reports that M3.1 Flash used 180 million tokens during its index evaluation, against a median of 99 million. The comparison is for that evaluation, not a universal measure of usage, but it shows why total token consumption matters alongside the price per token.

Performance claims also depend on the task and test setup. MiniMax’s published charts show M3.1 Flash ahead on Automation-Bench, while DeepSeek V4.1 Flash 1017 leads on DeepSWE and Terminal-Bench among the models listed. Artificial Analysis puts M3.1 Flash behind several open-weight leaders on its Intelligence Index. These results do not settle which model is best for a particular workflow; they give buyers reasons to test against their own tasks and costs. The missing newsroom scores mean this account cannot support a conclusion about M3.1 Flash’s reliability at producing accurate articles.

From Space Bunny to M3.1 Flash

Before its public identification, the model appeared anonymously on OpenRouter as Space Bunny Alpha. According to the source account, it went live on September 23, led OpenRouter’s usage ranking for 10 days and processed 56.7 trillion tokens on that platform before being withdrawn on October 5. MiniMax says the model processed more than 200 trillion tokens overall. The company identified it as an early version of M3.1 Flash on October 19.

MiniMax’s comparison charts include Automation-Bench, DeepSWE, Terminal-Bench, OSWorld V2, Agents’ Last Exam and HLE without tools. The company selected the rival models, and the account says several listed competitors were not the newest versions available. Those scores are self-reported, and the source says they have not been reproduced independently. MiniMax’s claimed security result, including 82% on CyberGym, also comes from an internal evaluation, according to the account.

“More than 200 tokens per second in everyday use, with a median above 170.”

— MiniMax

Weights, Testing and Data Questions

The weights had not been released at the time described; MiniMax’s stated date was October 23, 2026. The source provides no license or official parameter count. A separate outlet reportedly gave a parameter estimate, but MiniMax had not confirmed it, so that figure is not established here. It is also unclear whether the weights would arrive on schedule or what license terms would apply.

The model’s speed and benchmark results remain dependent on test conditions. MiniMax’s charts are company-reported, and the newsroom account supplies its test method but not the scores or comparison table. That omission prevents a factual account of the test’s outcome. The source also says that, while Space Bunny Alpha was anonymous on OpenRouter, the provider could retain prompts but could not use them for training. MiniMax had not stated what happened to that stored data, leaving its handling unresolved.

The October 23 Weights Release

The next stated milestone is MiniMax’s promised October 23 weights release. Its publication would allow reviewers to check the model’s license, inspect the released configuration and run additional tests, though those outcomes are not guaranteed by the announcement. Independent replications of the company’s benchmark and security results would help clarify how broadly they apply.

For the newsroom test, the missing English and German score table is needed before readers can judge the results or compare M3.1 Flash with the other models tested. Until then, the supported conclusion is limited: the model is released at a low listed token price, its measured speed in one independent index was below MiniMax’s claim, and its newsroom-test performance remains undisclosed in the source material.

Key Questions

When did MiniMax release M3.1 Flash?

MiniMax released M3.1 Flash on October 19, 2026, according to the source account. The model was available through the MiniMax API, M Plan, MiniMax Code and OpenRouter.

How much does M3.1 Flash cost?

The listed price is $0.15 per million input tokens and $0.60 per million output tokens. Cached input is listed at $0.015 per million tokens. These are token prices, not a guaranteed cost for completing a task.

How fast is the model?

MiniMax claims everyday speed above 200 tokens per second and a median above 170. Artificial Analysis measured 120 tokens per second in its evaluation, so the reported figures differ.

Are M3.1 Flash’s weights available?

Not in the status described by the source. MiniMax said the weights would arrive on October 23, 2026; the source does not confirm that they were released or provide a license.

What did the newsroom test find?

The described test covered 24 articles based on 12 sources in English and German, scored by two judges. The source material does not include the scores, so it does not establish how M3.1 Flash performed.

Source: ThorstenMeyerAI.com