MiniMax M3.1 Flash: How It Did In Our Open-Model Newsroom Test

🔍 Read the full analysis: MiniMax M3.1 Flash: How It Did In Our Open-Model Newsroom Test on ThorstenMeyerAI.com

Before you orderOffer from Amazon

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

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

MiniMax released M3.1 Flash on October 19, presenting it as a faster, lower-cost model for agent tasks and pricing it at $0.15 per million input tokens and $0.60 per million output tokens. A report by Thorsten Meyer says the model was compared with open-weight competitors on an independent index and in a 24-article newsroom test. The available report details include the independent scores and test setup, but not the newsroom results, so that test cannot support a conclusion about its writing quality.

MiniMax offers M3.1 Flash through its API, M Plan subscription, MiniMax Code and OpenRouter. OpenRouter made it free for its first 72 hours. The listed prices are half the per-token rates of MiniMax M3; cached input is priced at $0.015 per million tokens. The model has a stated one-million-token context window, adjustable reasoning effort and an OpenRouter output limit of 524,288 tokens.

MiniMax says the model produces more than 200 tokens per second in everyday use, with a median above 170, and describes it as nearly three times faster than M3. Those are company claims. Artificial Analysis measured 120 output tokens per second and 1.08 seconds to first token in its own evaluation. The index gave M3.1 Flash an Intelligence Index score of 43, below the listed scores for DeepSeek V4.1 Flash 1017 and GLM-5.3, both at 45.

The newsroom evaluation used 12 real news sources in English and German, producing 24 articles. Two judges, GPT-6.1 Sol and Claude Opus 5.5, were to score each article from 1 to 5 for source fidelity, invented facts, clarity, usefulness, structure and polish. The provided source says the model ran through a free OpenRouter endpoint with a 32,000-token output limit, but its results table is cut off. No scores or comparative ranking from that test can be reported from the available material.

Price, Speed and Editorial Reliability

The launch puts a lower-priced model into a crowded field where cost per token is only one part of the calculation. Artificial Analysis found that M3.1 Flash used 180 million tokens to complete its index evaluation, compared with a median of 99 million. That difference does not by itself establish the cost of a particular newsroom workflow, but it suggests that a low token price may not translate directly into the lowest total cost when models use different numbers of tokens.

For publishers and other teams using models to turn source documents into drafts, the missing newsroom scores matter. A model can be fast and inexpensive yet still require substantial human review if it adds unsupported details or misses key information. The described test evaluates those risks, but without its results readers cannot judge how M3.1 Flash performed against alternatives on that task.

From Space Bunny to M3.1 Flash

Before its public naming, the model operated on OpenRouter as Space Bunny Alpha. According to Meyer’s report, it appeared on September 23, led OpenRouter’s usage ranking for 10 days and processed 56.7 trillion tokens on the platform before it was removed on October 5. MiniMax says total usage across services exceeded 200 trillion tokens. These are reported usage figures, not an independent measure of model quality.

MiniMax also says M3.1 Flash uses “more aggressive sparsity” and costs about 70% less per agent task than M3. That task-cost comparison is the company’s claim; the source does not provide enough detail to independently assess its assumptions. MiniMax’s published benchmark charts also compare the model with a selected group of rivals, and the report notes that several listed competitors were older versions. Benchmark results in those charts are self-reported, while Artificial Analysis supplies a separate evaluation.

““more aggressive sparsity””

— MiniMax

Weights and Test Results Pending

The model’s weights had not been released when the report was written, and MiniMax had not disclosed a license or official parameter count. A reported parameter estimate of 428 billion total and 23 billion active was not confirmed by the company and is not established by the supplied material. The report also says that, during the model’s anonymous OpenRouter period, the provider could retain prompts but could not use them for training; MiniMax had not clarified what happened to that data.

Several performance claims remain unverified independently. MiniMax’s benchmark scores and its reported 82% result on CyberGym come from company evaluations, according to the report. The supplied newsroom test description lacks its results table, so its scores, language-by-language outcomes and comparison with other models remain unknown. The source also does not establish whether Artificial Analysis’s measurements and MiniMax’s speed claims used comparable settings.

Watch for Weights and Scores

The next stated milestone is MiniMax’s promised October 23 weights release. Once available, the license and model configuration will be relevant to organizations deciding whether they can run or adapt the model themselves. Readers should also look for the missing newsroom results before drawing conclusions about its reliability for publishing work, along with independent replication of the company’s benchmark and security claims.

Key Questions

How much does MiniMax M3.1 Flash cost?

MiniMax lists $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. OpenRouter offered the model free for its first 72 hours after release.

How did it perform in the newsroom test?

The report describes a test of 24 articles based on 12 sources, written in English and German and scored by two judges. The results table is missing from the supplied material, so the scores and ranking cannot be confirmed here.

Is M3.1 Flash open-weight?

Not yet according to the report. The weights had not been released, and Artificial Analysis listed it as proprietary at the time. MiniMax said the weights would arrive on October 23, 2026; no license was provided in the source.

Is it faster than DeepSeek V4.1 Flash 1017?

MiniMax claims more than 200 tokens per second in everyday use. Artificial Analysis measured M3.1 Flash at 120 tokens per second, while its listed figure for DeepSeek V4.1 Flash 1017 was 211. These figures come from the index’s measurements, not a universal speed guarantee.

Source: ThorstenMeyerAI.com