MiniMax M3.1 Flash And Open Models: What Our Newsroom Test Found

🔍 Read the full analysis: MiniMax M3.1 Flash And Open Models: What Our Newsroom Test Found on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• 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, pricing it at $0.15 per million input tokens and $0.60 per million output tokens, half the listed per-token price of its M3 predecessor. A report by Thorsten Meyer says it also tested the model on 24 real news articles, but the supplied source ends before reporting the scores, leaving the newsroom comparison unresolved.

MiniMax describes M3.1 Flash as a fast, low-cost model for agent tasks. It says the model can exceed 200 tokens per second in everyday use, with a median above 170, and is nearly three times faster than M3. MiniMax also claims agent tasks cost about 70% less than with M3. These are company claims; the report cites a separate Artificial Analysis measurement of 120 tokens per second and 1.08 seconds to first token.

The model has a one-million-token context window and adjustable reasoning effort. Cached input is priced at $0.015 per million tokens. At publication, OpenRouter offered the model free for its first 72 hours. The weights had not been released, and MiniMax said they would arrive October 23. The company had not supplied an official parameter count or license.

The report also describes a newsroom test using 12 real sources in English and German, producing 24 articles under a production prompt. Two judges, GPT-6.1 Sol and Claude Opus 5.5, were to score each article for source fidelity, invented facts, clarity, usefulness, structure and polish. But the supplied material provides no scores or model-by-model results. It therefore supports describing the test design, not declaring M3.1 Flash a newsroom winner or loser.

Price Claims Meet Editorial Accuracy

The launch matters to teams evaluating AI models for agent workflows and content production because lower token prices can reduce operating costs. But price alone does not show whether a model is cheaper to use for a completed task: token consumption, output quality and the amount of human correction also matter. Artificial Analysis found M3.1 Flash used 180 million tokens to complete its index, compared with a 99 million median. That is a raw comparison of the reported totals, not a direct measure of cost for every use case.

For newsrooms, the missing benchmark scores are particularly important. A model that writes quickly or cheaply is not necessarily reliable when it must preserve source details and avoid invention. The reported test was designed to measure those editorial qualities, but without its results readers cannot judge the trade-off or compare the model with alternatives. MiniMax’s launch claims and an independent index offer useful evidence, but neither substitutes for the absent newsroom findings.

From Space Bunny to M3.1 Flash

According to the report, the model appeared anonymously on OpenRouter as Space Bunny Alpha on September 23. It ranked as the platform’s most-used model for ten days, processed 56.7 trillion tokens there, and was removed on October 5. MiniMax later identified it as an early version of M3.1 Flash. The report says MiniMax counted more than 200 trillion tokens overall, a company-reported figure.

During the anonymous period, the provider was allowed to retain prompts but not use them for training, according to the source. MiniMax had not clarified what happened to that data. The report also compares company-published benchmark charts with outside measurements. MiniMax’s charts show mixed results against selected rivals, while Artificial Analysis gives M3.1 Flash an Intelligence Index score of 43, below several open-weight models listed in the report. The weights’ unreleased status led Artificial Analysis to classify it as proprietary at that time.

“The company says the weights “will arrive on Oct 23.””

— MiniMax

Newsroom Scores Are Missing

The supplied source material does not include the 24-article test results, the scores from either judge, or comparisons with the other models said to be in the evaluation. It is therefore unknown from this material how M3.1 Flash performed on source fidelity, factual invention, clarity or the other criteria. The test ran through a free OpenRouter endpoint with a 32,000-token output limit, details that could also matter when interpreting any eventual results.

Other points remain unsettled. MiniMax had not released the weights or stated their license or official parameter count. The source says an outlet reported a parameter estimate, but MiniMax had not confirmed it. MiniMax’s benchmark and security figures, including its reported CyberGym result, were company evaluations and had not been independently reproduced in the report. The provider’s handling of prompts retained during the Space Bunny period was also not specified.

Weights and Test Results Awaited

The next stated milestone is the October 23 weights release, which would allow researchers and developers to examine the model under the terms of its eventual license. The source does not confirm that the release occurred, so its availability and licensing remain to be checked.

A complete assessment also requires the newsroom benchmark’s missing scores and the comparison models’ results. Readers should look for the full table, the exact evaluation conditions and any follow-up on retained prompts. Until those details are available, the reported launch price and independent measurements can be assessed, but the model’s newsroom performance cannot be concluded from the supplied report.

Key Questions

When did MiniMax release M3.1 Flash?

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

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.

Did the newsroom test show that M3.1 Flash performed well?

The supplied source describes a 24-article evaluation but does not provide its scores or comparison results. It does not establish whether the model performed well or poorly in that test.

Are M3.1 Flash’s weights available?

They had not been released when the report was published. MiniMax said they would arrive on October 23; the supplied material does not confirm whether that happened or specify a license.

How did its measured speed compare with MiniMax’s claim?

MiniMax claimed more than 200 tokens per second in everyday use. The report cites Artificial Analysis measuring 120 tokens per second and 1.08 seconds to first token, using a separate evaluation.

Source: ThorstenMeyerAI.com