🔍 Read the full analysis: MiniMax M3.1 Flash: Open-Weight Leaderboard Results Meet Our Newsroom Findings on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
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. An independent index scored it below several open-weight competitors, while a newsroom test covered 24 English- and German-language articles; the source report does not provide that test’s scores. MiniMax says the model’s weights will be released October 23, but its license and other technical details remain undisclosed.
MiniMax released M3.1 Flash on October 19 at prices of $0.15 per million input tokens and $0.60 per million output tokens, presenting it as a fast, lower-cost model for agent tasks. Early comparisons are mixed: Artificial Analysis scored it below several open-weight rivals on its Intelligence Index, while a newsroom report describes a 24-article test but does not include its results in the available material. MiniMax says the model’s weights are due October 23.
The model is available through the MiniMax API, M Plan subscription, MiniMax Code and OpenRouter. OpenRouter offered access free for the first 72 hours. The listed price for cached input is $0.015 per million tokens; OpenRouter lists the earlier M3 model at twice M3.1 Flash’s input and output rates. MiniMax says Flash costs about 70% less per agent task than M3, but that is the company’s claim rather than an independently verified comparison.
MiniMax advertises a one-million-token context window, adjustable reasoning effort 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 tokens per second and 1.08 seconds to the first token, below the company’s speed claim. The measurements may use different conditions, and the source does not provide a controlled comparison resolving the gap.
In Artificial Analysis’ index, M3.1 Flash scored 43 on the Intelligence Index. The listed scores were 45 for DeepSeek V4.1 Flash 1017 and GLM-5.3, and 46 for Xiaomi’s MiMo-V2.6-Pro. The index also measured M3.1 Flash using 180 million tokens to complete its evaluations, compared with a 99-million-token median. That result is relevant to total usage costs: a low price per token does not by itself establish a lower cost for a completed task.
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
Price Versus Measured Performance
M3.1 Flash’s pricing makes it relevant to developers choosing models for high-volume agent workflows, where token costs and response speed can affect operating budgets. But the early evidence does not support treating its low per-token price as proof that it is the cheapest option for completed work. Artificial Analysis’ token-use result suggests that how much a model generates can change the cost calculation.
The reported comparisons also show why benchmark claims need careful reading. MiniMax’s published charts put M3.1 Flash ahead on Automation-Bench, while DeepSeek V4.1 Flash 1017 scored higher on DeepSWE and Terminal-Bench. The source report says MiniMax selected the competitors and that some were older versions. These are company-reported results, not independent replications, so they offer a limited basis for ranking models across tasks.
For newsroom and other factual writing, the central question is not a benchmark score alone but whether the model preserves source details and avoids invention. The described newsroom evaluation uses real articles and two judges, which addresses that practical question. However, without the scores or findings, it cannot support a conclusion about M3.1 Flash’s accuracy or suitability for publication.
As an affiliate, we earn on qualifying purchases.
From Space Bunny to Release
Before its public identification, the model ran anonymously on OpenRouter as Space Bunny Alpha. According to the source report, it appeared on September 23, led OpenRouter’s usage ranking for 10 days and processed 56.7 trillion tokens on that platform before being removed on October 5. MiniMax says it processed more than 200 trillion tokens overall; that total is the company’s figure.
The anonymous period raises a data question for users who sent prompts to the service. The report says the provider was allowed to retain prompts during that period but not use them for training. MiniMax has not said what happened to that retained data, according to the source. The report also says the model is currently treated as proprietary by Artificial Analysis because its weights have not yet been released.
MiniMax’s benchmark charts, as transcribed by CellCog in the source report, include results for Automation-Bench, DeepSWE, Terminal-Bench, OSWorld V2, Agents’ Last Exam and HLE. MiniMax reported a 62.2 score on Automation-Bench, ahead of DeepSeek V4.1 Flash 1017’s 60.5. DeepSeek scored 77.0 on DeepSWE and 42.3 on Terminal-Bench, above MiniMax’s 73.7 and 38.1. MiniMax’s charts did not report every rival’s result in every category, and its security figures, including an 82% CyberGym score, are described as internal evaluations.
““more aggressive sparsity””
— MiniMax
As an affiliate, we earn on qualifying purchases.
Weights, Data and Test Results
The source report says MiniMax had not released the model weights as of October 20, 2026, and the promised October 23 date had not yet arrived. It gives no confirmed license or official parameter count. A separate outlet reportedly estimated 428 billion total parameters and 23 billion active parameters, but MiniMax had not confirmed those figures, so they should not be treated as established specifications.
The newsroom test is described as running M3.1 Flash on a free OpenRouter endpoint, with a 32,000-token output limit. It used 12 sources in English and German, producing 24 articles, which were scored by GPT-6.1 Sol and Claude Opus 5.5. The available source material ends before providing the results table. No newsroom score or ranking can be confirmed from the information supplied.
Other open questions include the conditions behind MiniMax’s speed claims, whether its benchmark results can be reproduced independently, and what happened to prompts retained during the anonymous OpenRouter period. MiniMax’s claim that the model ranked first among Chinese models on Agents’ Last Exam also needs qualification: the report notes that MiniMax’s chart omitted DeepSeek from that row, while DeepSeek’s model card reports a different score that may reflect a different setup.
As an affiliate, we earn on qualifying purchases.
Awaiting the Weight Release
The next stated milestone is MiniMax’s planned October 23 weights release. Once available, the license and published model files will determine how developers can inspect, run or adapt the model. MiniMax had not supplied those details in the source report, so the release date and terms should be checked against the company’s announcement.
Further independent testing could clarify how M3.1 Flash compares with newer competing models under matched conditions, including speed, token use, task completion costs and factual accuracy. The newsroom results would also help establish whether its benchmark performance translates to source-grounded article writing; those scores are not present in the material available here.
As an affiliate, we earn on qualifying purchases.
Key Questions
When was MiniMax M3.1 Flash released?
MiniMax released M3.1 Flash on October 19, 2026, through its API, M Plan subscription, MiniMax Code and OpenRouter, according to the source report.
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 than M3 has not been independently established in the source material.
Are the model weights available?
Not as of the source report dated October 20, 2026. MiniMax said the weights would arrive October 23; the report provides no confirmed license or official parameter count.
How did M3.1 Flash compare with other models?
Artificial Analysis gave it an Intelligence Index score of 43, below the listed scores of 45 for DeepSeek V4.1 Flash 1017 and GLM-5.3 and 46 for MiMo-V2.6-Pro. MiniMax’s own benchmarks showed a mixed picture, with different models leading different tasks.
What did the newsroom test find?
The source describes a test of 24 articles across English and German, evaluated by two judges against the original sources. The available material does not include the scores or findings, so it does not establish how well M3.1 Flash performed on that test.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
