What Sets MiniMax M3.1 Flash Apart In Our Newsroom Test?

🔍 Read the full analysis: What Sets MiniMax M3.1 Flash Apart In Our 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, pricing it at $0.15 per million input tokens and $0.60 per million output tokens and pitching it as a fast, low-cost model for agent work. A newsroom review described a test of the model on 24 real articles in English and German, but the supplied source material does not include the scores, leaving its performance on the core writing task unreported.

The model is available through the MiniMax API, M Plan, MiniMax Code and OpenRouter. OpenRouter offered it free for its first 72 hours. MiniMax says the price per token is half that of M3 and that an agent task costs about 70% less; the latter is a company claim, not an independently reported result. Cached input is listed at $0.015 per million tokens.

MiniMax claims everyday generation speeds above 200 tokens per second, with a median above 170, and says the model is nearly three times faster than M3. Artificial Analysis measured 120 tokens per second and 1.08 seconds to the first token, below the company’s speed claim. The model has a stated one-million-token context window; MiniMax has disclosed only that its design uses “more aggressive sparsity.”

The newsroom evaluation used 12 real sources, each turned into an English and a German article, with the same production prompt used for other models. Two judges, GPT-6.1 Sol and Claude Opus 5.5, rated source fidelity, invented facts, clarity, usefulness, structure and polish on a five-point scale. The review says it ran the model through a free OpenRouter endpoint with a 32,000-token output limit. However, the supplied article cuts off before the score table, so no newsroom result or ranking can be confirmed.

Price Cuts Meet Writing Evidence

M3.1 Flash’s lower listed price could matter to developers running many model calls, particularly for agent workflows. But price per token is only one part of the cost: the review reports that Artificial Analysis counted 180 million tokens to complete its index, compared with a 99-million-token median. That index result suggests a model’s token consumption can affect total expense even when its unit price is low; it does not establish how much a particular newsroom or agent deployment would spend.

For publishers, the key question is whether lower cost and claimed speed come without sacrificing faithful reporting. The described newsroom test was designed to examine that question, including whether the model invents facts. Without its scores, readers cannot tell whether M3.1 Flash performed well, poorly or differently across English and German. The independent index and newsroom test address separate tasks and should not be treated as substitutes for one another.

From Space Bunny to M3.1

Before its public identification, the model operated as Space Bunny Alpha on OpenRouter from September 23 until it was removed on October 5. The source says it led the platform’s usage ranking for 10 days and processed 56.7 trillion tokens there. MiniMax puts overall use above 200 trillion tokens. These figures are attributed to the review’s account of platform and company data; they do not measure writing quality.

MiniMax’s published benchmark comparison includes results on Automation-Bench, DeepSWE, Terminal-Bench, OSWorld V2, Agents’ Last Exam and HLE. The review notes that the company selected the competitors and that many comparisons used older versions of rival models. In the listed results, DeepSeek V4.1 Flash 1017 scored above M3.1 Flash on DeepSWE and Terminal-Bench, while M3.1 Flash led on Automation-Bench. MiniMax’s benchmark figures are self-reported, and the review says its security results, including an 82% CyberGym score, have not been independently reproduced.

Artificial Analysis gave M3.1 Flash an Intelligence Index score of 43, below several open-weight models listed in the review. The index classified M3.1 Flash as proprietary because its weights had not yet been released. MiniMax said the weights would arrive October 23, but the source does not provide confirmation that they were subsequently published or specify a license.

““nearly three times faster” than M3”

— MiniMax

Missing Scores and Data Details

The supplied source material does not show the newsroom benchmark’s results table. It names the models whose scores were to be compared, but provides no scores for M3.1 Flash or the reference models. Claims about its newsroom accuracy, language differences or overall ranking would therefore go beyond the evidence available here.

It is also unclear whether the promised weights were released on October 23, what license would apply, or what the model’s parameter count is. The review says an outlet reported 428 billion total parameters and 23 billion active parameters, but MiniMax had not confirmed those figures. The source also says that during the anonymous Space Bunny period, the provider could retain prompts but could not use them for training; MiniMax had not clarified what happened to that retained data.

Benchmark comparisons have limits: MiniMax selected its rivals, some versions were older, and its chart did not include DeepSeek in one Agents’ Last Exam row. The source notes that DeepSeek’s model card reports a different score, possibly under a different setup. Those figures do not establish a directly comparable result.

Weights and Full Test Results

The next stated milestone was MiniMax’s promised October 23 weights release. Confirmation of publication, the accompanying license and any official parameter details would help establish whether the model is available for independent inspection. MiniMax could also provide information about retained prompts from the Space Bunny period.

For readers evaluating the newsroom claims, the missing score table is the immediate gap. Publication of the complete results, including scores for both judges and the English and German runs, would allow a fair comparison with the other tested models. Until then, the available record supports conclusions about the launch price, company speed claims, independent index measurements and test design—but not a verdict on newsroom writing performance.

Key Questions

When did MiniMax release M3.1 Flash?

MiniMax released M3.1 Flash on October 19, 2026. The source lists it on 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, $0.60 per million output tokens and $0.015 per million cached input tokens. MiniMax says its per-token price is half M3’s.

Did M3.1 Flash do well in the newsroom test?

The available source describes a test of 24 articles but does not include the score table. Its newsroom performance cannot be determined from the supplied results.

How fast is M3.1 Flash?

MiniMax claims everyday speeds above 200 tokens per second. Artificial Analysis measured 120 tokens per second and 1.08 seconds to the first token, so measured speed was below the company’s stated figure.

Are the model weights available?

MiniMax said the weights would arrive on October 23, 2026. The source does not confirm whether they were released or state what license would cover them.

Source: ThorstenMeyerAI.com