How Our Newsroom Test Challenged The Leaderboard View Of MiniMax M3.1 Flash

🔍 Read the full analysis: How Our Newsroom Test Challenged The Leaderboard View Of MiniMax M3.1 Flash 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, pricing it at $0.15 per million input tokens and $0.60 per million output tokens, as ThorstenMeyerAI.com published an assessment comparing the company’s claims with independent measurements and its own newsroom test. The report describes a 24-article evaluation, but the source material supplied for this article ends before its results table, so the model’s newsroom score and comparative standing cannot be confirmed here.

MiniMax is marketing M3.1 Flash as a low-cost, fast model for agent tasks. The company says it is nearly three times faster than M3, with everyday speeds above 200 tokens per second and a median above 170. Those figures are company claims. Artificial Analysis measured 120 tokens per second and 1.08 seconds to first token in its own evaluation, a lower speed result under that index’s testing conditions.

The model is available through the MiniMax API, M Plan, MiniMax Code and OpenRouter. Its listed price is half M3’s per-token rate, according to the report; cached input is priced at $0.015 per million tokens. MiniMax says it costs about 70% less per agent task, but that figure is also a company claim and depends on the task and token use. The model has a one-million-token context window. Its weights were not available at publication, and MiniMax said they would arrive on October 23.

The report also compared MiniMax’s published benchmark charts with an index run by Artificial Analysis. M3.1 Flash scored 43 on the Intelligence Index, below several models listed in the report, including DeepSeek V4.1 Flash 1017 at 45. Artificial Analysis measured the MiniMax model using 180 million tokens to complete its index, compared with a median of 99 million. That is a measure of token use in the index, not a result from the newsroom test.

Price Claims Meet Real-Task Testing

The report asks a practical question that leaderboard scores alone do not answer: can a model produce a reliable article from source material, without adding unsupported details? For newsrooms and other buyers, accuracy, clarity and editing burden can matter as much as speed or price per token. A model that uses substantially more tokens may also cost more on a completed task than its low headline rate suggests.

However, the information supplied here does not include the newsroom test’s scores, the rival models’ results, or the judges’ findings. It is not possible to say whether M3.1 Flash passed the test, how it ranked, or whether it was better or worse than alternatives. The reported design of the evaluation is useful context, but it is not a substitute for its missing results.

The benchmark comparisons also have limits. MiniMax selected competitors for its own charts, and the report says some were older releases. Artificial Analysis provides a separate measurement, but its index evaluates a defined set of tasks rather than every newsroom workflow. Readers should treat these as different kinds of evidence, not a single definitive ranking.

From Space Bunny to M3.1 Flash

Before the public release, the model operated anonymously on OpenRouter as Space Bunny Alpha. According to the report, it appeared on September 23, topped the platform’s usage ranking for 10 days, and was removed on October 5 after processing 56.7 trillion tokens there. MiniMax says it processed more than 200 trillion tokens overall. Those totals describe use, not a measured quality score.

MiniMax identified Space Bunny Alpha as an early version of M3.1 Flash on October 19. During the anonymous period, the provider could retain prompts but was not allowed to train on them, according to the report. MiniMax had not clarified what happened to that retained data. That question matters to users who submitted confidential material during the preview.

The report describes MiniMax’s benchmark figures as self-reported and says its security results, including an 82% CyberGym score, came from an internal evaluation that had not been independently reproduced. It also notes that the model’s reported parameter count was unconfirmed by MiniMax. Artificial Analysis classified it as proprietary while its weights remained unreleased.

““nearly three times faster” than M3”

— MiniMax

Missing Scores Limit the Comparison

The source material provided for this article describes the newsroom test setup but cuts off immediately after the model-and-language results table begins. It gives no article scores, rival model results, overall ranking, or judge comments. The test’s outcome is therefore unknown, and no conclusion about M3.1 Flash’s newsroom quality can be drawn from the available text.

Other open questions include whether the weights will be released on the promised date, what license will apply, and whether independent evaluators will reproduce MiniMax’s benchmark and security claims. MiniMax had not disclosed an official parameter count. It had also not explained how prompts retained during the Space Bunny period were handled.

Reported speed and cost figures also need to be read with their methods in mind. MiniMax’s claims and Artificial Analysis’s measurements may reflect different settings and workloads. The supplied material does not establish that either figure predicts performance or cost for every agent task or newsroom assignment.

Weights and Test Results to Watch

The next dated milestone in the report is MiniMax’s promised October 23 weights release. Their availability, license and technical details would help determine whether developers can inspect or run the model outside MiniMax’s hosted services. The source does not confirm that the release has occurred.

For readers evaluating the newsroom claim, the missing results table is the key item to obtain. Scores for both languages, results for the comparison models and the judges’ assessments would show whether the test supports any conclusion about fidelity, fabricated details or writing quality. Until those data are available, the most supportable account is limited to the test design and the separately reported price and index measurements.

Key Questions

When did MiniMax release M3.1 Flash?

MiniMax released M3.1 Flash on October 19, 2026, according to the source report.

What 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.

How did it score in the newsroom test?

The supplied source material does not include the newsroom scores or comparison results. It describes a test of 24 articles, evaluated by two judges, but its outcome cannot be confirmed from the text available.

Are MiniMax’s speed figures independently verified?

MiniMax claims everyday speeds above 200 tokens per second and a median above 170. Artificial Analysis measured 120 tokens per second in its evaluation. These are results from different sources and testing conditions.

When were the model weights expected?

MiniMax said the weights would arrive on October 23, 2026. The source report, published October 20, does not confirm whether they were subsequently released.

Source: ThorstenMeyerAI.com