MiniMax M3.1 Flash: Testing Its Newsroom Performance Against Open Models

🔍 Read the full analysis: MiniMax M3.1 Flash: Testing Its Newsroom Performance Against Open Models 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, pricing the model at $0.15 per million input tokens and $0.60 per million output tokens, and positioning it as a low-cost option for agent tasks. A review published by ThorstenMeyerAI.com compared MiniMax’s claims with Artificial Analysis measurements and a newsroom evaluation of 24 articles; the supplied results show a gap between the company’s speed claims and independent measurements, but do not include the newsroom scores.

The release is available through the MiniMax API, M Plan subscription, MiniMax Code and OpenRouter. OpenRouter offered access free for the first 72 hours, according to the review. MiniMax lists cached input at $0.015 per million tokens. Its predecessor, M3, is listed by OpenRouter at $0.30 per million input tokens and $1.20 per million output tokens, making M3.1 Flash half the price per token on those measures.

MiniMax says the model can exceed 200 tokens per second in everyday use, with a median above 170, and calls it nearly three times faster than M3. Artificial Analysis measured 120 tokens per second and 1.08 seconds to first token. The index gave M3.1 Flash an Intelligence Index score of 43, compared with 45 for DeepSeek V4.1 Flash 1017 and GLM-5.3, and 46 for Xiaomi’s MiMo-V2.6-Pro. Those figures are measurements from the index, not a direct result from the newsroom test.

The review’s newsroom setup used 12 real news sources in English and German, producing 24 articles with the same production prompt. Two judges, GPT-6.1 Sol and Claude Opus 5.5, scored source fidelity, invented facts, clarity, usefulness, structure and polish on a five-point scale. The review says M3.1 Flash ran through OpenRouter with a 32,000-token output limit. However, the source material provided for this article ends before the score table, so it does not establish how the model performed against other models in that test.

The Price-Speed Trade-Off

M3.1 Flash’s price could make it attractive for applications that generate many responses, including agent workflows and publishing pipelines. But cost per token is not the same as cost per completed task: Artificial Analysis reports that the model used 180 million tokens to complete its index evaluation, compared with a 99 million-token median. That does not by itself establish what a newsroom or business would pay, but it points to a possible trade-off between low token prices and the amount of output needed.

For newsrooms, benchmark scores are only part of the decision. A model must preserve source details, avoid inventing information and produce usable copy consistently. The described test addresses those questions, but without its results, readers cannot judge whether M3.1 Flash’s low price comes with acceptable editorial performance. The distinction matters when a model is considered for real publishing work rather than coding or general benchmark tasks.

From Space Bunny to M3.1

Before its public release, the model was offered anonymously on OpenRouter as Space Bunny Alpha. The review says it appeared on September 23, led OpenRouter’s usage ranking for ten days and processed 56.7 trillion tokens there before being removed on October 5. MiniMax says the model processed more than 200 trillion tokens overall. These are platform and company figures reported by the review; the underlying measurement details are not included in the supplied material.

MiniMax describes the model’s design only as using “more aggressive sparsity” and gives a context window of one million tokens. Its weights have not yet been released, so Artificial Analysis lists it as proprietary for now. MiniMax says the weights will arrive October 23, but no license or official parameter count has been announced. A separate outlet’s reported parameter figures are not confirmed by MiniMax and are not treated here as established facts.

MiniMax’s own benchmark charts, transcribed by CellCog according to the review, show a mixed picture. M3.1 Flash leads DeepSeek V4.1 Flash 1017 on Automation-Bench by 1.7 points, while DeepSeek scores higher on DeepSWE and Terminal-Bench. The review notes that MiniMax chose the comparison models and that several were older versions. It also says the company’s security results, including an 82% CyberGym score, come from internal evaluation and have not been independently reproduced.

““more aggressive sparsity””

— MiniMax

Missing Scores and Model Details

The supplied source material does not contain the newsroom benchmark’s score table, so the 24-article results and model comparisons cannot be reported. It is also unclear how the judges’ individual ratings differed, whether the test included human review, or how the model handled errors across particular article types. The review describes the method but not the outcome in the material provided.

Other open questions include the weights’ final release and license, the model’s official parameter count, and what happened to prompts submitted during the anonymous Space Bunny period. The review says the provider could retain prompts but could not use them for training; MiniMax has not said what happens to that retained data. Artificial Analysis’s speed and token-use measurements also differ from MiniMax’s claims, and the supplied account does not explain the testing conditions behind the company’s speed figures.

Weights Due October 23

The next stated milestone is MiniMax’s planned October 23 release of the weights. Their publication would allow users to inspect the accompanying license and, depending on the release terms, run the model outside MiniMax’s hosted services. Until that happens, it remains listed as proprietary by Artificial Analysis.

Readers should also look for the missing newsroom score table and any independent replication of MiniMax’s benchmarks and security claims. For now, the available evidence supports a lower listed token price and a measured speed below the company’s stated everyday-use figure, but it does not answer whether M3.1 Flash produces reliable newsroom copy or lowers overall costs in practice.

Key Questions

When did MiniMax release M3.1 Flash?

MiniMax released M3.1 Flash on October 19, 2026, through its API, M Plan subscription, MiniMax Code and OpenRouter, according to the review.

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.

Was M3.1 Flash faster than MiniMax claimed?

Artificial Analysis measured 120 tokens per second, while MiniMax said everyday use could exceed 200 tokens per second, with a median above 170. The supplied material does not explain the different testing conditions.

What did the newsroom test find?

The review describes a test of 24 articles based on 12 news sources, scored by two independent judges. The supplied source ends before the results table, so the scores and model rankings are not available here.

Are the model weights available?

Not yet, according to the review. MiniMax said the weights were due on October 23, 2026; no license or official parameter count was disclosed in the supplied material.

Source: ThorstenMeyerAI.com