TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
MiniMax has released M3.1 Flash, a model it says is faster and cheaper to use for agentic tasks, coding and computer operation. The company lists API prices of $0.15 per million input tokens and $0.60 per million output tokens; its performance and speed figures are company-reported, and model weights are not yet available.
MiniMax has released M3.1 Flash, a model aimed at coding, computer-use and other multi-step AI tasks, and says it is available through the MiniMax API, M Plan and MiniMax Code. The company lists token pricing from $0.15 per million input tokens and says the model cuts agentic-task usage costs by 70% compared with its predecessor, though that comparison and the performance results are company claims rather than independently verified findings in the supplied announcement.
MiniMax describes M3.1 Flash as a next-generation model for tasks in which an AI system must plan, use tools and carry out work across multiple steps. The company says it improves on a broad set of benchmarks covering enterprise workflows, software development, computer operation and cybersecurity. Its announcement does not provide independent evaluations or explain all benchmark testing conditions, so the scores should be read as figures reported by MiniMax.
The company reports a score of 76.2 on BankerToolBench and a result of 73.7% on DeepSWE v1.1, which MiniMax describes as a benchmark for complex software tasks. For computer use, it reports 61.7 points on OSWorld V2 and 29.6 points on Agents’ Last Exam, adding that the model ranks first among Chinese models on both. MiniMax also reports an 82% score on CyberGym. The source does not provide comparison tables, evaluation dates or details sufficient to independently assess those rankings.
MiniMax says output speeds can reach more than 200 tokens per second, with median speeds above 170 tokens per second. Its listed prices per one million tokens are $0.15 for input, $0.60 for output and $0.015 for cached input. The company says overall usage costs for agentic tasks are 70% below those of its predecessor, but does not specify the task mix or calculation behind that comparison in the announcement. Model weights, which would allow access outside the listed services, are scheduled for a later release.
Lower Costs for Multi-Step AI Work
The release targets a practical concern for organizations using AI agents: a system that can complete longer workflows may also consume more tokens and incur higher bills. MiniMax is positioning M3.1 Flash as a lower-cost option for work involving software, business applications and computer interfaces, with its stated pricing and speed intended to make repeated tool use more affordable.
Those benefits are not yet established beyond MiniMax’s own account. Actual costs depend on how much a task uses the model, whether input can be cached, and how reliably it completes work without retries or human correction. Likewise, benchmark scores do not by themselves show how the model performs in a particular company’s systems or whether it can safely handle sensitive tasks. Buyers will need to test the model against their own workloads and compare total task costs, not just per-token rates.
The cybersecurity claims merit similar care. MiniMax says its internal experts evaluated code auditing, vulnerability patching, penetration testing, malware analysis and reverse engineering. It also reports a CyberGym result. The announcement does not include the underlying security evaluation or establish that the model is suitable for unsupervised security operations. The stated capabilities are relevant to security teams, but need separate validation in controlled settings.
As an affiliate, we earn on qualifying purchases.
Where M3.1 Flash Is Available
MiniMax says users can access M3.1 Flash through three routes: its API, M Plan subscription and MiniMax Code. API users can pay according to the published per-token rates, while the company also points to its subscription and coding product. The announcement does not detail whether every feature, limit or model setting is identical across all three access routes.
The company also gives a backstory for the name “Space Bunny.” It says an early version of M3.1 Flash operated anonymously and processed more than 200 trillion tokens before the official rollout. MiniMax does not provide dates, deployment details or a method for verifying that figure in the supplied material, so it should be treated as the company’s description of the model’s earlier use.
MiniMax frames Flash as an effort to balance capability with efficiency and says a separate Pro model, intended to push intelligence further, will arrive later. It does not give a release date for Pro. The announcement also says M3.1 Flash’s model weights will be released later, leaving the current offering focused on access through MiniMax’s products and services.
““Today, we officially roll out MiniMax M3.1 Flash.””
— MiniMax
Benchmark Claims Need Independent Checks
The supplied announcement is from MiniMax, and the performance, ranking, speed and cost comparisons are claims made by the company. It does not link to detailed benchmark results or explain test configurations, model versions used for comparisons, sample sizes or how costs were calculated. Those details matter when comparing models, since results can vary with prompts, tools, hardware and evaluation rules.
It is also unclear when model weights will be released, what licensing terms will apply, and whether access limits or prices differ across the API, M Plan and MiniMax Code. MiniMax has not provided a date for its planned Pro model in the source material. The announcement does not establish how the model performs in independent production deployments or what safeguards are available for higher-risk cybersecurity use.
Weights and Pro Release Still Pending
Readers can currently access M3.1 Flash through the MiniMax API, M Plan and MiniMax Code, according to the company. The next stated product milestone is a later release of model weights, but MiniMax has not announced a date or licensing terms. The company also says its Pro model is coming later without giving a schedule.
Further information to watch for includes detailed benchmark documentation, independent evaluations and clarification of the 70% cost comparison. Those disclosures would help developers and organizations determine whether the model’s reported speed, task performance and pricing translate into savings and dependable results for their own use cases.
Key Questions
What is MiniMax M3.1 Flash?
MiniMax describes M3.1 Flash as a model for agentic tasks, coding, computer use and cybersecurity-related work. The company says it is designed to handle multi-step tasks and tool use.
How much does M3.1 Flash cost?
MiniMax lists prices per one million tokens of $0.15 for input, $0.60 for output and $0.015 for cached input. The announcement does not provide a total cost estimate for a typical task.
Where can users access the model?
MiniMax says M3.1 Flash is available through its API, M Plan and MiniMax Code. The company has not specified whether all access routes offer the same features or limits.
Are the benchmark results independently verified?
The supplied release reports the benchmark scores, but does not provide independent verification or detailed test documentation. The figures should be treated as MiniMax-reported results unless supported by separate evaluations.
When will the model weights be released?
MiniMax says the weights will be released later, but has not provided a date or licensing terms in the announcement.
Source: hn
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
