MiniMax M3.1 Flash: From Leaderboard Underdog To Newsroom Test Standout

🔍 Read the full analysis: MiniMax M3.1 Flash: From Leaderboard Underdog To Newsroom Test Standout on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

MiniMax released M3.1 Flash on October 19 at a low per-token price, positioning it as a fast model for agent work. A review by ThorstenMeyerAI.com found the model’s performance depends on the test: it led one company-selected automation benchmark, while Artificial Analysis scored it behind several open-weight competitors. The review also ran 24 news articles through the model, but the source material supplied for this report does not include the resulting scores.

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. Those listed input and output rates are half the prices shown for its M3 predecessor. MiniMax says the new model costs about 70% less per agent task, but that figure is the company’s claim, not an independently established result. The model is available through the MiniMax API, M Plan, MiniMax Code and OpenRouter; OpenRouter offered access free for the first 72 hours.

MiniMax advertises a one-million-token context window and everyday speeds above 200 tokens per second, with a median above 170. The company describes the design change only as “more aggressive sparsity.” Artificial Analysis measured 120 output tokens per second and 1.08 seconds to first token, below the company’s speed claim and below its measured rate for DeepSeek V4.1 Flash 1017. The two figures come from different testing contexts, so they should not be treated as directly equivalent measurements.

In MiniMax’s published comparison, M3.1 Flash scored 62.2 on Automation-Bench, ahead of DeepSeek V4.1 Flash 1017 at 60.5. DeepSeek led the listed models on DeepSWE and Terminal-Bench. MiniMax’s charts did not report every rival’s score on every test, and the company selected the comparisons. The review says these benchmark results are self-reported and have not yet been independently reproduced.

Price Meets Real-World Performance

The release adds another low-priced option for developers building AI agents and automated workflows, where repeated model calls can make usage costs a practical constraint. But a low token price does not by itself establish lower total costs: Artificial Analysis found M3.1 Flash used 180 million tokens to complete its index, compared with a median of 99 million. That is a raw token-use comparison in the index, not a universal measure of how much the model will cost for a particular task.

The different results also show why model rankings need to be read alongside their methods. MiniMax’s selected benchmarks suggest strengths in automation, while the independent index places the model below several open-weight systems on its Intelligence Index. For newsroom use, the review’s test design is directly relevant: it asks whether a model can turn real source material into accurate, useful articles. However, without the reported scores, readers cannot assess how M3.1 Flash performed against the other tested models.

From Anonymous Traffic to Launch

Before the public release, the model reportedly appeared on OpenRouter under the name Space Bunny Alpha. According to the review, it went live anonymously on September 23, led the platform’s usage ranking for 10 days and processed 56.7 trillion tokens there before being removed on October 5. MiniMax says the model processed more than 200 trillion tokens overall; that total is the company’s figure.

The review says users’ prompts could be retained by the provider during the anonymous OpenRouter period, although they could not be used for training under the stated arrangement. MiniMax had not clarified what happened to that data. M3.1 Flash’s weights were also still unreleased at the time of the report, so Artificial Analysis listed it as proprietary. MiniMax said weights would arrive on October 23; it had not announced a license or confirmed a parameter count in the material supplied.

““more aggressive sparsity””

— MiniMax

Scores and Data Still Missing

The supplied source ends after introducing a results table for the newsroom test; it does not provide the table’s scores or identify how M3.1 Flash compared with the other models. The review describes a test of 24 articles, created from 12 real news sources in English and German and judged by GPT-6.1 Sol and Claude Opus 5.5, but the outcome cannot be independently summarized from the material available here.

Other open questions include whether the promised weights will be released on schedule and under what license, and whether independent researchers will reproduce MiniMax’s benchmark claims. The source also does not say what happened to prompts retained during the model’s anonymous OpenRouter period. A separate outlet’s reported parameter count is not confirmed by MiniMax and is excluded from this account.

Weights and Independent Rechecks

The next stated milestone is MiniMax’s planned October 23 weight release. Its publication would allow researchers to inspect the model and evaluate it outside the hosted API, subject to the license MiniMax sets. Readers should look for confirmation of the release, its terms and any further technical details, including a company-confirmed parameter count.

Further independent testing could clarify how the model performs on coding, agent tasks and source-grounded writing, and whether its speed and token use hold across other workloads. The newsroom comparison would also be more useful with its missing scores and model-by-model results published. Until then, the available evidence supports a measured conclusion: M3.1 Flash is inexpensive by listed token price and competitive on some reported tests, but its standing across tasks remains unsettled.

Key Questions

When did MiniMax release M3.1 Flash?

MiniMax released M3.1 Flash on October 19, 2026, according to the review. The company said the model weights would arrive on October 23.

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. MiniMax’s claim that it costs about 70% less per agent task is a company comparison, not an independently verified result in the supplied material.

Did M3.1 Flash beat DeepSeek V4.1 Flash?

Not across the reported tests. MiniMax’s charts show M3.1 Flash ahead on Automation-Bench, while DeepSeek V4.1 Flash 1017 scored higher on DeepSWE and Terminal-Bench. The figures are from MiniMax’s own comparison and have not been independently reproduced, according to the review.

How did M3.1 Flash score in the newsroom test?

The review says it tested the model on 24 English- and German-language articles, using two judges and a source-fidelity rubric. The supplied source does not include the results table, so the model’s scores and ranking cannot be reported.

Are the model weights available?

They had not been released at the time covered by the report. MiniMax said the weights were scheduled for October 23; the source does not confirm whether that release occurred or specify a license.

Source: ThorstenMeyerAI.com