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 launched M3.1 Flash, a model it says improves performance on agentic tasks, coding and computer use while cutting usage costs by 70% compared with its predecessor. The company lists API prices starting at $0.15 per million input tokens; independent verification of its benchmark and speed claims is not included in the announcement.
MiniMax has launched M3.1 Flash, a new AI model available through the company’s API, M Plan subscription and MiniMax Code product. The company says it is designed for agentic tasks, coding and computer use, with 70% lower usage costs than its predecessor; its model weights are not yet available and are scheduled for a later release.
MiniMax’s announcement reports results on several evaluation benchmarks. It says M3.1 Flash scored 76.2 on BankerToolBench and 73.7% on DeepSWE v1.1, which tests complex software engineering tasks. For computer-use evaluations, MiniMax reports scores of 61.7 on OSWorld V2 and 29.6 on Agents’ Last Exam, adding that the model ranked first among Chinese models on those two benchmarks. The announcement does not provide a full methodology or independent evaluation alongside these figures.
For cybersecurity, MiniMax reports an 82% score on CyberGym and describes internal assessments of code auditing, vulnerability patching, penetration testing, malware analysis and reverse engineering. These are company-reported results, including findings from its own security experts; they should not be read as independent validation. The announcement characterizes M3.1 Flash as stronger across these areas but does not give a detailed comparison table covering all competing models.
MiniMax lists prices per million tokens of $0.15 for input, $0.60 for output and $0.015 for cached input. It says everyday output can reach more than 200 tokens per second, with median speeds above 170 tokens per second, and attributes efficiency improvements to more aggressive sparsity. Those speed figures are company claims, and the report does not specify the hardware, workload or measurement conditions behind them.
Lower Costs for Agent Workflows
The launch matters to developers and businesses considering AI systems that carry out multi-step work, use software tools or operate across applications. Lower per-token prices can reduce the cost of frequent model calls, while faster responses may make interactive workflows more practical. Whether those advantages translate into lower total project costs depends on the task, the amount of input and output, and how reliably the model completes work without human correction.
MiniMax’s benchmark claims are relevant because they cover several tasks associated with deployed agents, including software changes and computer operation. But scores alone do not establish how a model will perform in a customer’s environment. Buyers will need to test accuracy, consistency, security and total cost against their own tasks and alternatives. The published pricing gives users a concrete starting point for that comparison, while the performance figures remain claims from the model provider.
As an affiliate, we earn on qualifying purchases.
Flash Model and Planned Weights
MiniMax describes M3.1 Flash as a next-generation Flash model focused on balancing capability and inference efficiency. The company says an early version operated anonymously under the name Space Bunny and processed more than 200 trillion tokens. That figure and the model’s anonymous operation are reported by MiniMax; the announcement does not provide further details about when or where this use occurred.
The model is currently offered through MiniMax API, M Plan and MiniMax Code. MiniMax says it will release model weights later, but does not give a date or specify the release terms. It also says a separate Pro model is expected later, describing it as focused on pushing intelligence capabilities. No launch date or technical details for Pro are provided in the report.
““Today, we officially roll out MiniMax M3.1 Flash.””
— MiniMax
As an affiliate, we earn on qualifying purchases.
Independent Testing Still Pending
The source material is MiniMax’s own launch report. It does not include third-party test results, detailed benchmark protocols, comparison data for every named competitor, or independent confirmation of the reported speed and cost advantages. The reported figures therefore describe MiniMax’s claims, not a complete external assessment of the model.
Several practical details remain unspecified: the date for the weight release, the license or access conditions that would apply, and the hardware and tasks used to measure token speed. It is also unclear how the model’s performance varies across different languages, domains and longer production workflows. The announcement does not provide enough detail to establish how its cybersecurity capabilities should be assessed outside the company’s own evaluations.
As an affiliate, we earn on qualifying purchases.
Weights and Pro Await Dates
Developers can access M3.1 Flash now through the MiniMax API, M Plan and MiniMax Code, according to the company. The next stated milestone is a later release of the model weights, but MiniMax has not announced a timetable or terms. The company also says its Pro model will arrive later, with no date given.
For users weighing adoption, the immediate next step is to compare the published rates with actual task costs and test the model on representative workflows. Further information from MiniMax on evaluation methods, weight availability and product timelines would help clarify how the launch claims translate into real-world use.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is MiniMax M3.1 Flash?
M3.1 Flash is a MiniMax AI model marketed for agentic tasks, coding, computer use and cybersecurity work. MiniMax describes it as a next-generation Flash model focused on capability and inference efficiency.
How much does M3.1 Flash cost?
MiniMax lists prices per million tokens of $0.15 for input, $0.60 for output and $0.015 for cached input. Actual costs depend on usage and the mix of input, output and cached tokens.
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 yet released the model weights.
Are the benchmark and speed claims independently verified?
The supplied announcement reports the benchmark scores and inference speeds, but does not include independent verification or detailed test conditions. The figures should be treated as MiniMax-reported results.
When will MiniMax release the weights?
MiniMax says the weights will be released later, but has not provided a date, license or other release terms.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
