MiniMax M3.1 Flash: Strong Newsroom Results Despite A Lower Leaderboard Position

🔍 Read the full analysis: MiniMax M3.1 Flash: Strong Newsroom Results Despite A Lower Leaderboard Position on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get hardware and tech essentials 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. A review published by ThorstenMeyerAI.com on October 20 reports strong performance in its newsroom test of 24 articles, while an Artificial Analysis index score of 43 places the model behind several open-weight competitors.

The model is available through the MiniMax API, M Plan subscription, MiniMax Code and OpenRouter. OpenRouter listed it free for its first 72 hours. MiniMax describes M3.1 Flash as a fast, inexpensive option for agent work and says its per-token price is half that of M3. It also claims the model costs about 70% less per agent task, runs at more than 200 tokens per second in everyday use and has a median speed above 170 tokens per second. Those speed and cost figures are MiniMax’s claims, not results established by the independent measurements cited in the review.

Artificial Analysis measured M3.1 Flash at 120 output tokens per second and 1.08 seconds to the first token. Its Intelligence Index score was 43. The index lists 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. Artificial Analysis also found the model used 180 million tokens to complete its index evaluations, compared with a listed median of 99 million. The review says that higher token use can reduce the practical benefit of a low per-token price.

ThorstenMeyerAI.com tested the model on 24 articles based on 12 news sources, producing an English and German article from each source. Two judges, GPT-6.1 Sol and Claude Opus 5.5, assessed source fidelity, invented facts, clarity, usefulness, structure and polish. The source describes the newsroom results as strong, but the supplied material ends before its results table. It therefore provides no scores or model-by-model comparisons for that test.

Low Price Meets Uneven Performance

The results point to a practical trade-off for newsrooms and developers: low token prices and good task performance do not by themselves establish that a model is the most efficient choice. M3.1 Flash’s newsroom evaluation is relevant because it tests work grounded in real source material, including whether a model adds unsupported facts. But without the table of scores, readers cannot judge how large the reported advantage was or how it compared with other models under the same conditions.

The independent index gives a different measure. M3.1 Flash’s score of 43 trails several listed open-weight models, and its measured speed is below MiniMax’s stated everyday-use figure. The index also found comparatively high token consumption. Together, the measures caution against judging the model on its list price alone; actual cost and usefulness depend on the task, the output required and the number of tokens used.

For news organizations, errors in a generated article can carry consequences beyond cost or speed. The test’s focus on source fidelity and invented facts addresses an editorial concern, but a 24-article evaluation is still a limited sample. Its findings do not establish how the model will perform across other subjects, languages, workflows or production conditions.

From Space Bunny to M3.1 Flash

Before the public release, the model appeared anonymously on OpenRouter as Space Bunny Alpha. According to ThorstenMeyerAI.com, it was available from September 23 until it was pulled on October 5, led OpenRouter’s usage ranking for 10 days and processed 56.7 trillion tokens on that platform. MiniMax says the model processed more than 200 trillion tokens overall. These figures describe different scopes and are attributed to the outlet and company, respectively.

The review says users of Space Bunny Alpha were told the provider could retain prompts but could not use them for training. MiniMax has not explained what happens to that retained data, according to the source. The model has a stated one-million-token context window, adjustable reasoning effort and an OpenRouter output allowance of up to 524,288 tokens. Its weights had not been released at the time of the review, and MiniMax had not announced a license or official parameter count.

In MiniMax’s published benchmark charts, M3.1 Flash scored 62.2 on Automation-Bench, compared with 60.5 for DeepSeek V4.1 Flash 1017. DeepSeek scored higher on DeepSWE, at 77.0 versus 73.7, and Terminal-Bench, at 42.3 versus 38.1. The review notes that these are company-reported benchmark results, with MiniMax selecting the comparison models. It also says the charts omit DeepSeek’s score in the Agents’ Last Exam comparison, and that independent reproduction of the reported results was not available.

“More than 200 tokens per second in everyday use, with a median above 170.”

— MiniMax

Missing Scores and Unreleased Weights

The supplied source material does not include the newsroom results table referenced in the review. It describes M3.1 Flash’s performance as strong, but does not show its scores, the results for other models in that test or the margin between them. The precise basis for that characterization therefore cannot be checked from the material available here.

The model weights were not yet available as of the October 20 review; MiniMax had said they would arrive on October 23. The company had also not published an official parameter count or license. A parameter estimate reported by another outlet was not confirmed by MiniMax and is not treated here as established fact. It is also unclear how MiniMax will handle prompts retained during the model’s anonymous OpenRouter period.

MiniMax’s benchmark and security claims remain company-reported in the cited material. The source says the security results, including an 82% CyberGym result, came from an internal evaluation and had not been independently reproduced. Differences in benchmark setups may also affect comparisons, including the reported Agents’ Last Exam figures.

Weights and Independent Testing

The next stated milestone was MiniMax’s planned October 23 weights release. Its publication would allow developers and researchers to inspect the model under the terms of the license, if one is provided, and could clarify whether it can be classified as open-weight. The release was still in the future when the review appeared.

Readers evaluating the model should look for the missing newsroom scores and any further testing that reports methods, comparison models and results in full. Independent replication of MiniMax’s benchmark claims, along with an explanation of how prompts from the Space Bunny period were handled, would address important gaps in the current account. Until then, the available evidence combines company claims, one independent index and a newsroom test whose detailed scores are absent from the supplied source.

Key Questions

What is MiniMax M3.1 Flash?

M3.1 Flash is a MiniMax model released October 19, 2026. The company presents it as a low-cost, fast model for agent tasks and lists a one-million-token context window.

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. OpenRouter offered it free for its first 72 hours after release, according to the review.

How did it perform in the newsroom test?

ThorstenMeyerAI.com described the results as strong after testing 24 articles from 12 sources in English and German. The supplied source does not include the scores or comparison table, so the scale of the performance cannot be verified here.

Why is its leaderboard position lower than its newsroom results suggest?

The measures evaluate different things. Artificial Analysis gave M3.1 Flash an Intelligence Index score of 43, below several models it lists, while the newsroom test assessed article writing against real source material. The detailed newsroom scores are missing from the source provided, preventing a direct comparison.

When were the model weights expected?

MiniMax said the weights would arrive on October 23, 2026. They had not been released at the time of the October 20 review, and the source does not confirm whether the planned release occurred.

Source: ThorstenMeyerAI.com