🔍 Read the full analysis: Why MiniMax M3.1 Flash Performed Differently In Our Newsroom Test on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
MiniMax released M3.1 Flash on October 19, 2026, pricing it at $0.15 per million input tokens and $0.60 per million output tokens. The source reports that independent benchmark results lag some open-weight rivals, but it provides no newsroom scores, so the model’s performance on the 24-article test cannot be established from the available material.
MiniMax released M3.1 Flash on October 19, pricing the model below its predecessor and promoting it for agent work, while a review from ThorstenMeyerAI.com found that its results vary across company claims, an independent index and an in-house newsroom test. The supplied review material describes the newsroom test’s 24 articles and judging process but does not include its scores, leaving the central newsroom comparison unconfirmed 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 standard rates are half the listed input and output prices for M3, according to the source. MiniMax says Flash is nearly three times faster than M3 and costs about 70% less per agent task. Those performance and task-cost figures are company claims, not results established by the newsroom test.
The review says Artificial Analysis gave M3.1 Flash an Intelligence Index score of 43. It measured output speed at 120 tokens per second and time to first token at 1.08 seconds, below MiniMax’s claim of more than 200 tokens per second in everyday use. The index comparison listed DeepSeek V4.1 Flash 1017 at 45 and GLM-5.3 at 45. Artificial Analysis also recorded 180 million tokens used to complete its evaluation, compared with a median of 99 million across the evaluated models.
For the newsroom test, the review says researchers used 12 real news sources, each turned into English and German articles for a total of 24 outputs. Two judges, GPT-6.1 Sol and Claude Opus 5.5, scored each article from 1 to 5 on source fidelity, invented facts, clarity, usefulness, structure and polish. The model ran through OpenRouter’s free endpoint with a 32,000-token output limit. The supplied source stops before the results table, so it does not show M3.1 Flash’s scores or how it ranked against other models.
Infographic 01 · MiniMax M3.1 Flash
A step behind on the index. Ahead of every open model in our newsroom.
An independent intelligence index puts MiniMax M3.1 Flash a step behind the open-weight leaders. Writing real news from real sources, it beat all of them.
Independent index
Artificial Analysis Intelligence Index, 10 evaluations. Bar length on a 0–50 scale.
Artificial Analysis lists M3.1 Flash as proprietary until its weights are released.
Our newsroom test
Mean score, English articles, 1–5 by GPT-6.1 Sol and Claude Opus 5.5 with the source in context. Bar length on a 0–5 scale.
In German, M3.1 Flash (3.36) also leads every open model, but GPT-6 Luna is clearly ahead (3.62).
Sources: Artificial Analysis, MiniMax-M3.1-Flash and open-weights models (20 Oct 2026); Thorsten Meyer AI newsroom benchmark, 12 sources × English + German, run 20 Oct 2026.
Powered by Thorsten Meyer AI
Why Newsroom Accuracy Matters
For publishers, the practical question is not just whether a model generates text quickly or costs less per token. It is whether the final article stays faithful to its source, avoids invented details and is clear enough to use. Those qualities affect editorial reliability and the time required for human review.
The source’s index measurements suggest a trade-off worth checking: M3.1 Flash’s listed price is low, but Artificial Analysis recorded more tokens used in its evaluation than the index median. A cheaper token rate does not by itself establish a lower total cost for a task. The review’s newsroom design could help assess writing quality, but without its reported scores, readers cannot tell whether M3.1 Flash performed better or worse than the tested alternatives.
AI language model performance benchmark
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Anonymous Preview to Release
Before the public release, the model appeared on OpenRouter under the name Space Bunny Alpha. The review says it went live anonymously on September 23, led OpenRouter’s usage ranking for 10 days and was removed on October 5, after processing 56.7 trillion tokens on that platform. MiniMax separately counted more than 200 trillion tokens overall, according to the source.
MiniMax says the model has a one-million-token context window and adjustable reasoning effort. The company had not released the weights at the time covered by the review, but said they would arrive October 23. It had also not published an official parameter count or license. The review notes that an outside report gave a parameter estimate, but MiniMax had not confirmed it.
MiniMax’s own benchmark charts show M3.1 Flash ahead on Automation-Bench, while DeepSeek V4.1 Flash 1017 scored higher on DeepSWE and Terminal-Bench. The review cautions that these company-published results are self-reported and that the models included in the comparisons were selected by MiniMax. Artificial Analysis supplies a separate comparison using its own evaluation suite.
newsroom AI content generation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Missing Newsroom Scores
The central uncertainty is the newsroom result itself: the source gives the test method but not the score table. It is not possible from the supplied material to say how M3.1 Flash scored in English or German, whether it generated unsupported facts, or how it compared with other models tested under the same prompt.
Other details also remained unresolved in the review. MiniMax had not confirmed the model’s parameter count or published its weights and license. The source says that during the anonymous Space Bunny period, the provider could retain prompts but was not allowed to train on them; it says MiniMax had not explained what happened to that data. The review also says company benchmark and security figures had not been independently reproduced.
As an affiliate, we earn on qualifying purchases.
Weights and Results to Watch
The next stated release milestone was October 23, 2026, when MiniMax said it planned to publish the model weights. Their release, along with a license and confirmed technical specifications, would allow users to assess how openly the model can be examined and deployed. The supplied material does not confirm whether that release occurred.
For the newsroom comparison, the missing English and German scores are needed before drawing a conclusion about editorial performance. Until those figures and the judge assessments are available, the review supports conclusions about the test setup and the reported index measurements, but not a verdict on how M3.1 Flash performed in the newsroom.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is MiniMax M3.1 Flash?
It is a MiniMax model released on October 19, 2026, marketed for agent work. The listed price is $0.15 per million input tokens and $0.60 per million output tokens.
How was the newsroom test conducted?
The review says it used 12 real news sources in English and German, producing 24 articles. Two judges scored each output on a 1-to-5 scale across source fidelity, invented facts, clarity, usefulness, structure and polish.
Did M3.1 Flash perform well in the newsroom test?
The supplied source does not show the test’s score table. It describes the method but does not provide the results, so its newsroom performance cannot be determined from this material.
Was M3.1 Flash open-weight at release?
No weights were available at the time covered by the review. MiniMax said they would arrive on October 23, 2026, but the supplied material does not confirm a later release or state a license.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
