🔍 Read the full analysis: MiniMax M3.1 Flash: It Trails The Open-Weight Leaderboard. In Our Newsroom Test, It Beat Every Open Model. on ThorstenMeyerAI.com
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
TL;DR
MiniMax released M3.1 Flash on October 19, 2026, at $0.15 per million input tokens and $0.60 per million output tokens. Artificial Analysis gave it an Intelligence Index score of 43, below several open-weight models, while a newsroom test covered 24 articles; the source does not provide that test’s scores. The model’s weights are promised for October 23, but have not yet been released.
MiniMax released M3.1 Flash on October 19, pricing the model below its predecessor and pitching it for fast, lower-cost agent work. But independent measurements place it behind several open-weight rivals on an intelligence index, and although a newsroom review says it tested the model on 24 real articles, the source material provided for this report does not include the newsroom scores.
The model is available through the MiniMax API, M Plan subscription, MiniMax Code and OpenRouter. Its listed price is $0.15 per million input tokens and $0.60 per million output tokens, with cached input priced at $0.015 per million tokens. OpenRouter offered it free for its first 72 hours. MiniMax says the price per token is half that of M3 and that agent tasks cost about 70% less; those savings are company claims, not independently verified results.
MiniMax advertises speeds above 200 tokens per second in everyday use, with a median above 170, and describes the model as nearly three times faster than M3. Artificial Analysis measured 120 tokens per second and 1.08 seconds to first token. The service supports a one-million-token context window, while OpenRouter lists a maximum output of 524,288 tokens. Reasoning effort can be adjusted.
The supplied newsroom review says its test used 12 real news sources in English and German, producing 24 articles with the same production prompt. GPT-6.1 Sol and Claude Opus 5.5 judged the articles on source fidelity, invented facts, clarity, usefulness, structure and polish. However, the source text ends before giving the results table. Its scores, and any ranking against other tested models, cannot be reported from the material available here.
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 low listed price may appeal to developers running agents at scale, but price per token is only part of the cost. Artificial Analysis says the model used 180 million tokens to complete its index evaluation, compared with a median of 99 million. That finding suggests token consumption could narrow the savings from a lower unit price, although the index’s token use does not by itself establish the total cost of a particular customer’s workload.
The index gave M3.1 Flash an Intelligence Index score of 43. That is below Xiaomi’s MiMo-V2.6-Pro at 46; DeepSeek V4.1 Flash 1017 and GLM-5.3 at 45; and Step 5 Preview and Kimi K3 at 44. The score makes the model a competitive low-cost option, but not the leader among the models listed. Developers weighing a switch need to compare performance, output volume, latency and task-specific results—not just headline token prices.
For newsrooms and other accuracy-sensitive users, the missing newsroom results are a material limitation. A model’s ability to produce fluent copy does not establish that it preserves source facts. Without the review’s scores or examples, readers cannot assess whether M3.1 Flash performed well on the test’s central concern: factual fidelity.
As an affiliate, we earn on qualifying purchases.
From Space Bunny to M3.1
Before its public release, the model operated anonymously on OpenRouter as Space Bunny Alpha. According to the review, it went live on September 23, led the platform’s usage ranking for 10 days, processed 56.7 trillion tokens there, and was removed on October 5. MiniMax says it processed more than 200 trillion tokens overall, though the source does not describe the accounting period or provide an independent verification of that figure.
MiniMax’s own benchmark charts compare M3.1 Flash with a selection of competitors. On those company-reported results, M3.1 Flash scored 62.2 on Automation-Bench, 1.7 points ahead of DeepSeek V4.1 Flash 1017. DeepSeek scored higher on DeepSWE, 77.0 to 73.7, and Terminal-Bench, 42.3 to 38.1. The charts did not report every competitor on every test, and the scores have not been independently reproduced according to the source.
MiniMax says M3.1 Flash ranks first among Chinese models on Agents’ Last Exam. The company’s chart does not include DeepSeek on that row, while DeepSeek’s model card reports a score of 35.4 for its 1017 model, potentially under a different setup. The comparison is therefore not conclusive. MiniMax’s reported security results, including 82% on CyberGym, also come from an internal evaluation.
““More aggressive sparsity.””
— MiniMax
Missing Scores and Unreleased Weights
The newsroom review’s source material does not include its results table, so the 24-article scores and comparisons are unknown. It is not possible to say from the available information whether the model’s English and German results differed, how often judges identified invented facts, or how it performed against other models in the same test.
The model weights have not yet been released, and MiniMax has not provided an official parameter count or license in the source material. A report elsewhere put the model at 428 billion total parameters and 23 billion active parameters, but MiniMax has not confirmed those figures. The anonymous OpenRouter period also raises an unanswered data-handling question: the provider could retain prompts, though not use them for training, and MiniMax has not said what happened to those stored prompts.
Independent replication of MiniMax’s benchmark and security claims is also not described. Results across benchmarks can depend on test versions, configurations and tool access; the source does not supply enough detail to reconcile all differences between the company charts and outside measurements.
The October 23 Weights Release
The next stated milestone is October 23, when MiniMax says it will release the weights. Their availability, accompanying license and any technical documentation will determine whether developers can inspect or run the model outside MiniMax’s hosted services. Until those details are published, “open-weight” comparisons should be treated cautiously: Artificial Analysis lists M3.1 Flash as proprietary at the time covered by the source.
Readers should also look for the missing newsroom results and for independent tests that reproduce the vendor’s speed, cost and benchmark claims. Those details would help establish whether the model’s lower token price translates into lower costs for real workloads, and whether it can meet accuracy requirements in publishing or other source-based tasks. The supplied material does not state when the newsroom scores will be released.
Key Questions
When did MiniMax release M3.1 Flash?
MiniMax released it on October 19, 2026. The model was offered through the MiniMax API, M Plan, MiniMax Code and OpenRouter.
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.
Does it lead the open-weight field?
Not on the Artificial Analysis results cited in the source. M3.1 Flash scored 43, below several listed models, including MiMo-V2.6-Pro at 46 and DeepSeek V4.1 Flash 1017 at 45.
What did the newsroom test find?
The review says it tested 24 articles based on 12 sources in English and German, with two judges evaluating them. The supplied source does not include the score table, so the test’s findings cannot be stated.
When will the model weights be available?
MiniMax said the weights will arrive on October 23, 2026. The source does not provide a license or confirm that the release has occurred.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
