MiniMax M3.1 Flash is available on Novita AI as a serverless model, four days after MiniMax released it on October 19, 2026. It targets coding agents, computer-use automation, and multi-step tool workflows that want lower cost per completed task: Novita AI lists $0.15 per 1M input tokens and $0.60 per 1M output tokens. You reach it through the OpenAI-compatible base URL https://api.novita.ai/openai with the model ID minimax/minimax-m3.1-flash.
Key Takeaways
- MiniMax M3.1 Flash is live on Novita AI as a serverless model with OpenAI-compatible access; the specs table below has the model ID, base URL, and endpoint families.
- Pricing is flat: $0.15 per 1M input tokens, $0.60 per 1M output tokens, and $0.015 per 1M cached input tokens.
- It keeps the M3 family’s 1M-token context window and allows up to 524,288 output tokens.
- MiniMax reports stronger agentic results than its predecessor: BankerToolBench 76.2, DeepSWE v1.1 73.7%, OSWorld V2 61.7, and roughly 70% lower usage cost for agentic tasks. These are MiniMax-reported figures, not Novita-run benchmarks.
MiniMax M3.1 Flash Specs and Pricing Summary
| Field | Details |
|---|---|
| Display name | MiniMax M3.1 Flash |
| Model ID | minimax/minimax-m3.1-flash |
| Base URL | https://api.novita.ai/openai |
| Endpoint family | Chat completions, Anthropic-compatible, Responses |
| Input modalities | Text, image, video |
| Output modality | Text |
| Context window | 1,000,000 tokens |
| Max output tokens | 524,288 tokens |
| Input price | $0.15 per 1M tokens |
| Cache-read price | $0.015 per 1M tokens |
| Output price | $0.60 per 1M tokens |
| Best fit | Cost-sensitive coding agents, computer-use automation, and chained tool workflows that need long context |
Specs and pricing from the Novita AI model page, checked October 23, 2026.
What Is MiniMax M3.1 Flash?
MiniMax describes M3.1 Flash as a next-generation Flash model for coding and agent scenarios, built around a 1-million-token context window. The launch post reports nearly three times faster online inference, about 70% lower usage cost than its predecessor, and output speeds of 200+ tokens per second with a median around 170+ TPS.
On Novita AI the model accepts text, image, and video input and returns text, so an agent can reason over screenshots, diagrams, or short recordings. It does not generate images or video, and MiniMax says the weights will be released later, which leaves the hosted API as the practical path today.
MiniMax M3.1 Flash API Access on Novita AI
Access follows the same pattern as other Novita-hosted LLMs. Point an OpenAI SDK client at https://api.novita.ai/openai, keep your existing key, and set the model string to minimax/minimax-m3.1-flash. The listing also exposes Anthropic-compatible and Responses endpoint families, so an Anthropic-style client can target the model without a new integration.
For request fields, streaming, image-content formatting, and tool-call syntax, use the current Novita AI chat completions reference. The MiniMax M3 API quick start builds the same request shape around the earlier M3 model ID; change the model string and the rest carries over.
Key Capabilities for Developers
The capabilities worth evaluating are the ones that change how much work an agent finishes before a human steps in.
| Signal | Score | What it measures | Developer takeaway |
|---|---|---|---|
| BankerToolBench | 76.2 | Multi-step planning, document consistency, financial compliance, chained tool calls | Test it on your own long-chain workflows before routing production traffic |
| DeepSWE v1.1 | 73.7% | Repository-scale feature work, issue fixing, and test validation | Compare it against your current coding model on the same repo tasks |
| OSWorld V2 | 61.7 | Cross-application computer operation | A candidate for GUI and browser automation agents |
| Agents’ Last Exam | 29.6 | Complex professional tasks with economic value | Treat it as a directional signal, not task-level certainty |
| CyberGym | 82% | Vulnerability discovery and validation | Relevant for security review or patch-validation agents |
Data from MiniMax’s M3.1 Flash release post, checked October 23, 2026. Scores are MiniMax-reported and should be validated on your own tasks.
BankerToolBench rewards consistency across documents, compliance rules, and chained tool calls, which is close to back-office automation spanning a CRM, a document store, and a billing API. DeepSWE v1.1 measures repository-scale changes verified by tests, and the 524,288-token output ceiling supports long patch sequences. OSWorld V2 and Agents’ Last Exam cover computer operation and economically valuable professional tasks; MiniMax says the model ranks first among Chinese models on both. CyberGym covers vulnerability discovery and patch validation.
Speed matters too: MiniMax reports 200+ TPS with a median near 170+, which helps streaming interfaces and agents that make many short calls. Cache reads at $0.015 per 1M tokens are one-tenth of the input rate, so a stable system prompt or tool schema gets the largest saving when it is served from cache.
When to Use MiniMax M3.1 Flash
- Coding agents over large repositories that need long context but not the largest model
- Tool-heavy automation across enterprise systems, where sessions chain CRM, document, and billing calls
- Screenshot- or recording-driven computer-use agents that return text decisions
- High-volume agent loops with a stable prefix, where cache reads lower the bill
When Not to Use MiniMax M3.1 Flash
- Image or video generation, since the output modality is text only
- Short, high-volume classification or chat, where a smaller model is usually cheaper
- Decisions that need independent benchmark evidence, because the scores are MiniMax-reported and the model is days old
- Self-hosting today, because MiniMax says the weights will be released later
If your prompts already fit M3’s limits and cost, switching only pays off when the speed and price difference shows up in your own measurements. The MiniMax M3 vs MiniMax M2.7 comparison covers the earlier models’ pricing tiers and context.
How MiniMax M3.1 Flash Fits Your API Workflow
An existing Novita AI integration needs one change: swap in the model ID from the specs table on your current base URL and key. Text-only request bodies should work unchanged; image and video content needs the formatting in the API reference.
Replay a handful of real agent traces instead of testing a generic prompt. Record solved-task rate, retries, latency, and cost per completed task, and keep the model ID in configuration so a workload that does not improve can move back to MiniMax M3 on Novita AI.
Conclusion
MiniMax M3.1 Flash is worth evaluating when an agent chains tool calls over long context and you want lower cost without dropping to a much smaller model. It is serverless and OpenAI-compatible on Novita AI, priced at $0.15/$0.60 per 1M input/output tokens with $0.015 cache reads, and keeps the M3 family’s 1M-token window. MiniMax’s reported agentic and computer-use scores are the reason to test it now; your own traces are the reason to adopt it.
FAQ
Is MiniMax M3.1 Flash available on Novita AI?
Yes. MiniMax M3.1 Flash is listed on Novita AI as a serverless model and is reachable through the OpenAI-compatible API with the model ID in the specs table above.
What is the model ID for MiniMax M3.1 Flash on Novita AI?
Use minimax/minimax-m3.1-flash on the OpenAI-compatible endpoint at https://api.novita.ai/openai. The MiniMax M3.1 Flash model page is the source of record if the listing changes.
How much does MiniMax M3.1 Flash cost on Novita AI?
Novita AI lists $0.15 per 1M input tokens, $0.015 per 1M cached input tokens, and $0.60 per 1M output tokens for MiniMax M3.1 Flash. The rate is flat, so budgets scale linearly with token volume rather than jumping at a context threshold.
Does MiniMax M3.1 Flash support function calling and structured output?
Yes. Novita AI lists function calling, structured output, reasoning, and Anthropic API compatibility for MiniMax M3.1 Flash, alongside serverless access.
How is MiniMax M3.1 Flash different from MiniMax M3?
MiniMax M3.1 Flash is the newer, faster, cheaper variant: MiniMax reports roughly three times faster online inference and about 70% lower usage cost on agentic tasks. MiniMax M3 remains the earlier option with the same 1M-token context. There is no Novita-run head-to-head benchmark, so compare both on your own traces if the decision is close.
Can MiniMax M3.1 Flash read images and video?
Yes, as input. Novita AI lists text, image, and video input with text-only output for MiniMax M3.1 Flash, so it can analyze or act on visual material but cannot generate images or video.
