🔍 Read the full analysis: MiniMax M3.1 Flash: A Test Of Open Models That Defied Its Leaderboard Rank on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
MiniMax released M3.1 Flash on October 19, 2026, pricing it at $0.15 per million input tokens and $0.60 per million output tokens. The source review reports strong performance on some company-selected benchmarks, but Artificial Analysis measured lower speed than MiniMax claims, and the supplied material does not include the newsroom test scores. Model weights are expected October 23, according to MiniMax.
MiniMax released M3.1 Flash on October 19, setting prices at $0.15 per million input tokens and $0.60 per million output tokens, as the company positions it as a fast, lower-cost model for agent work. A review published by ThorstenMeyerAI.com compared MiniMax’s claims with Artificial Analysis measurements and a newsroom test; the reported evidence is mixed, and the supplied review material does not include the newsroom test’s scores.
MiniMax says M3.1 Flash costs half as much per token as M3 and about 70% less per agent task. It also claims everyday output speed above 200 tokens per second, with a median above 170, describing the model as nearly three times faster than M3. The company has disclosed only that it uses “more aggressive sparsity” as a design change. These speed and task-cost figures are company claims, not independent measurements.
Artificial Analysis gives M3.1 Flash an Intelligence Index score of 43, below the 45 recorded for DeepSeek V4.1 Flash 1017 and GLM-5.3 in the cited comparison. It measured the model at 120 tokens per second, with 1.08 seconds to first token. The index test used 180 million tokens to complete, compared with a 99 million-token median. The review says that higher token use can reduce the benefit of a low per-token price, though total costs depend on the task and usage.
In MiniMax’s published benchmark charts, M3.1 Flash scored 62.2 on Automation-Bench, ahead of DeepSeek V4.1 Flash 1017’s 60.5. DeepSeek led on DeepSWE and Terminal-Bench. MiniMax’s charts did not report results for every rival on every test, and the review notes that MiniMax selected the comparison models. The company’s scores, including its security results, have not been independently reproduced in the material provided.
Infographic 01 · MiniMax M3.1 Flash
A step behind on the index. Ahead of every open model in our newsroom.
An independent intelligence index puts MiniMax M3.1 Flash a step behind the open-weight leaders. Writing real news from real sources, it beat all of them.
Independent index
Artificial Analysis Intelligence Index, 10 evaluations. Bar length on a 0–50 scale.
Artificial Analysis lists M3.1 Flash as proprietary until its weights are released.
Our newsroom test
Mean score, English articles, 1–5 by GPT-6.1 Sol and Claude Opus 5.5 with the source in context. Bar length on a 0–5 scale.
In German, M3.1 Flash (3.36) also leads every open model, but GPT-6 Luna is clearly ahead (3.62).
Sources: Artificial Analysis, MiniMax-M3.1-Flash and open-weights models (20 Oct 2026); Thorsten Meyer AI newsroom benchmark, 12 sources × English + German, run 20 Oct 2026.
Powered by Thorsten Meyer AI
Price and Performance Trade-Offs
The release adds another low-priced model aimed at agent workflows, where systems may make repeated model calls and process long inputs. Its listed input price is half M3’s $0.30 per million tokens, and its output price is half M3’s $1.20, according to the review. That makes the model’s actual token use and task performance relevant alongside its headline rates.
The measured gap between the company’s speed claim and Artificial Analysis’s result gives buyers a reason to compare models under their own workloads. The index measured 120 tokens per second, rather than the more than 200 tokens per second MiniMax says is typical. That does not settle how the model performs across different systems or tasks, but it cautions against treating a vendor’s speed figure as a universal result.
The review also highlights a distinction between benchmark ranking and newsroom quality. A model can score well on selected evaluations without necessarily producing reliable, accurate articles from source material. The provided source describes a newsroom test design but omits its results, so it cannot support a conclusion about M3.1 Flash’s performance on that task.
Popular right now
Samsung 43-Inch M70H Series 4K Mini LED Smart TV (2026)
Roborock RockNeo Q110H Wire-Free Robot Lawn Mower
KitchenAid 5.5-Quart Bowl-Lift Stand Mixer, Porcelain White
As an affiliate, we earn on qualifying purchases.
From Space Bunny to M3.1
Before its public identification, the model operated as Space Bunny Alpha on OpenRouter. The review says it appeared anonymously on September 23, led that 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. Those figures describe usage, not a measure of accuracy or quality.
MiniMax identified the model as an early version of M3.1 Flash on October 19 and made it available through its API, M Plan subscription, MiniMax Code and OpenRouter. OpenRouter offered it free for the first 72 hours, according to the review. The stated context window is one million tokens; the review reports an OpenRouter output limit of up to 524,288 tokens.
MiniMax says the model weights are due on October 23. Until they are released, Artificial Analysis lists M3.1 Flash as proprietary. The review says there is no official parameter count or license yet. It also notes that an outside report gave a parameter estimate, but MiniMax has not confirmed it, so the review excludes that figure.
““more aggressive sparsity””
— MiniMax
Unreleased Weights and Missing Scores
The source material does not provide the newsroom benchmark’s results. It describes a test of 24 articles—12 sources written in English and German—and says GPT-6.1 Sol and Claude Opus 5.5 judged the work, but the results table is absent. It is therefore not possible to report how M3.1 Flash scored for source fidelity, invented facts, clarity or other criteria, or to compare it with the other models in that test.
The weights, their license and the official parameter count were also still outstanding in the review. MiniMax has not confirmed how long it retained prompts sent to Space Bunny Alpha during its anonymous period. The review says the provider could keep prompts but could not train on them; what happened to those stored prompts remains unspecified. The benchmark results in MiniMax’s charts are company-reported, and the supplied material says independent reproduction has not yet taken place.
Comparisons also have limits: MiniMax selected the rivals in its charts, and the review says some listed competitors were older versions than models available at the time. Artificial Analysis’s measurements provide a separate comparison, but neither its index nor the company charts alone establish how the model will perform across every real-world workload.
Weights and Independent Testing
The next stated milestone is October 23, when MiniMax says it plans to release M3.1 Flash’s weights. Their availability, license and official parameter count will clarify whether developers can run and inspect the model outside the company’s hosted services. The release would also allow outside evaluators to test the same model under their own conditions.
For now, readers and potential users have the October 19 API release, MiniMax’s self-reported benchmarks and the Artificial Analysis measurements cited in the review. The newsroom test cannot be evaluated from the supplied account because its score table is missing. Any comparison of article-writing quality, or firm conclusion about total costs for agent tasks, will require those results or further testing with disclosed methods and scores.
Key Questions
When did MiniMax release M3.1 Flash?
MiniMax released M3.1 Flash on October 19, 2026, according to the source review.
How much does M3.1 Flash cost?
The review lists prices of $0.15 per million input tokens and $0.60 per million output tokens. Cached input is listed at $0.015 per million tokens.
Did independent testing confirm MiniMax’s speed claim?
No. MiniMax claimed everyday speed above 200 tokens per second. Artificial Analysis measured 120 tokens per second in its evaluation. These are different reported measurements, and the review does not establish how the model performs in every setting.
Are M3.1 Flash’s weights available?
Not according to the review published October 20. MiniMax said the weights were expected on October 23; their eventual availability and license are not confirmed in the supplied material.
What did the newsroom test find?
The supplied review describes a 24-article test scored by two judges but does not include the results table. No newsroom score or ranking can be confirmed from the material provided.
Source: ThorstenMeyerAI.com
