MiniMax M3.1 Flash And The Gap Between Leaderboard Rankings And Newsroom Results

🔍 Read the full analysis: MiniMax M3.1 Flash And The Gap Between Leaderboard Rankings And Newsroom Results on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

MiniMax released M3.1 Flash on October 19, pricing the model below its predecessor and presenting it as a fast, low-cost option for agent tasks. A review by ThorstenMeyerAI.com compared the company’s claims with an independent index and a newsroom test, finding that the available evidence does not support treating leaderboard rankings as a direct measure of performance on real reporting tasks.

MiniMax lists the API price at $0.15 per million input tokens and $0.60 per million output tokens, half the listed per-token rates for M3. Cached input costs $0.015 per million tokens. The company says M3.1 Flash is about 70% cheaper per agent task and can exceed 200 tokens per second in everyday use, with median speed above 170 tokens per second. Those speed and task-cost figures are company claims, not independently verified measurements in the supplied review.

The independent Artificial Analysis index gives M3.1 Flash an Intelligence Index score of 43. The review 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 measured M3.1 Flash at 120 output tokens per second and 1.08 seconds to first token, below MiniMax’s stated everyday speed. The model used 180 million tokens to complete the index, compared with a reported median of 99 million.

The review also describes a newsroom evaluation using 12 real news sources, each turned into English and German articles for 24 outputs. Two judges, GPT-6.1 Sol and Claude Opus 5.5, scored source fidelity, invented facts, clarity, usefulness, structure and polish. However, the supplied source material ends before the results table and gives no scores for M3.1 Flash or the comparison models. The newsroom outcome therefore cannot be reported from the information provided.

Why Newsroom Scores Matter

The comparison highlights a practical distinction: leaderboard performance and editorial reliability are different measures. A model can score well on selected coding or agent benchmarks without demonstrating that it can accurately turn source material into publishable news. For newsrooms, errors such as unsupported details or distorted claims can matter more than raw speed.

Price also depends on more than the listed rate per token. The review says M3.1 Flash consumed 180 million tokens on the independent index, against a 99 million-token median. That does not establish its cost on newsroom work, but it shows why token usage and task quality belong alongside per-token price when buyers compare models.

The available newsroom description gives readers a potentially useful evaluation design, but not the outcome. Without its scores, the report cannot establish whether M3.1 Flash performed better or worse than other models on the 24 articles, or whether its low price came with trade-offs in accuracy or editing effort.

From Space Bunny to M3.1

The model had already drawn attention before its public release. According to the report, it appeared anonymously on OpenRouter as Space Bunny Alpha on September 23, led the platform’s usage ranking for 10 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; that total is the company’s figure.

MiniMax’s published benchmark comparisons show M3.1 Flash ahead on Automation-Bench, scoring 62.2 against DeepSeek V4.1 Flash 1017’s 60.5. DeepSeek scored higher on DeepSWE, 77.0 to 73.7, and Terminal-Bench, 42.3 to 38.1. The review notes that MiniMax selected the models in its charts and that some comparisons use older versions of rival systems. It also says MiniMax’s security results are from an internal evaluation and have not been independently reproduced.

The release is not yet an open-weight launch. MiniMax says weights are due October 23, but the report says the company has not published a license or confirmed a parameter count. Artificial Analysis lists the model as proprietary while the weights remain unavailable.

“The model can exceed 200 tokens per second in everyday use.”

— MiniMax, as quoted in ThorstenMeyerAI.com’s report

The Missing Newsroom Results

The source material does not include the newsroom benchmark’s score table, so M3.1 Flash’s article-quality results remain unknown. It is not possible to compare the model’s English and German scores with rival systems or to draw a result from the test design alone.

Other details also remain unsettled. MiniMax has not confirmed the model’s parameter count, published the weights, or set out a license in the supplied report. The review says the anonymous OpenRouter provider could retain prompts during the Space Bunny period, although it could not use them for training; MiniMax has not said what happened to that data. The report also does not establish whether its internal security scores or company benchmark results have been reproduced by outside evaluators.

The October 23 Weights Release

The next stated milestone is MiniMax’s promised October 23 release of the weights. Readers and developers will be able to assess the license and inspect the released model if the company delivers them as described. Until then, it remains classified as proprietary by Artificial Analysis.

A fuller assessment of newsroom suitability requires the omitted test results or a repeatable release of the evaluation data and scoring. Independent checks of the model’s speed, task costs, security claims and prompt-data handling would also help clarify how its advertised price translates into practical use. No such outcomes are confirmed in the supplied report.

Key Questions

What is MiniMax M3.1 Flash?

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

How much does M3.1 Flash cost?

The listed API rates are $0.15 per million input tokens and $0.60 per million output tokens. Cached input is listed at $0.015 per million tokens.

Did M3.1 Flash win the newsroom test?

The supplied report describes a 24-article evaluation but does not include its score table. No newsroom winner or M3.1 Flash score can be confirmed from the available material.

How did it rank on the independent index?

Artificial Analysis gave M3.1 Flash an Intelligence Index score of 43. Several listed open-weight models scored higher, including MiMo-V2.6-Pro at 46 and DeepSeek V4.1 Flash 1017 at 45.

When will the model weights be available?

MiniMax said the weights would arrive on October 23, 2026. The supplied report says they had not yet been released and that no license had been announced.

Source: ThorstenMeyerAI.com