MiniMax's Flash-tier agentic workhorse — 1M-token context, tunable thinking depth, vision and video input, built for cheap multi-step coding and tool-use loops.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-m3-1-flash",
"messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
}'Overview
MiniMax M3.1 Flash is MiniMax's Flash-tier frontier LLM for agentic reasoning, coding and tool use, released October 19, 2026. It pairs a 1,000K-token context window with image and video input, web search, function calling, and a tunable thinking-depth setting, at a fraction of flagship-model pricing.
On Venice, MiniMax M3.1 Flash runs under the anonymized privacy tier: your prompts are not stored, profiled, or fed into a personal history — a sharp contrast to Big-Tech assistants that retain every query. You pay per token with no subscription lock-in, and you get the same agentic capabilities — function calling, vision, web search, reasoning — permissionlessly via the app or the Venice API. Note that this tier is anonymized rather than end-to-end encrypted, so it is private from Venice's storage practices, not cryptographically sealed in transit.
Agent quickstart
The API is OpenAI-compatible: change the base URL and the model id and existing client code works unchanged.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-m3-1-flash",
"stream": true,
"messages": [{ "role": "user", "content": "Draft the release note." }]
}'curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-m3-1-flash",
"messages": [{ "role": "user", "content": "Find the rate limits." }],
"tools": [{
"type": "function",
"function": {
"name": "search_docs",
"description": "Search the API documentation.",
"parameters": {
"type": "object",
"properties": { "query": { "type": "string" } },
"required": ["query"]
}
}
}],
"tool_choice": "auto"
}'import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["VENICE_API_KEY"],
base_url="https://api.venice.ai/api/v1",
)
resp = client.chat.completions.create(
model="minimax-m3-1-flash",
messages=[{"role": "user", "content": "Build without permission."}],
)
print(resp.choices[0].message.content)Specifications
Assessment
Use cases
Prompting
Set effort to "low" for routine agent steps and simple edits — it cuts latency and cost dramatically; reserve "max" (the default) for hard planning and multi-file changes.
Drop the entire codebase or document set into the 1M-token context instead of summarizing — the model is built for repo-scale input.
Structure prompts with stable prefixes (system prompt, tool schemas, reference files) so repeated calls hit the $0.02/1M cached-input rate.
Pass multiple screenshots or a screen recording as input when debugging UI or computer-use workflows — the model accepts multi-image and video input natively.
Define your tool schemas explicitly and let function calling drive the loop rather than describing tools in prose.
Ask for structured output (JSON schemas, diffs, checklists) — the model is tuned for structured task execution in agent harnesses.
Alternatives
| Model | Best for | Context | Open weights | Price (Venice) |
|---|---|---|---|---|
| MiniMax M3.1 Flash | Agentic coding & tool use at Flash-tier cost | 1M tokens | No | $0.19 in · $0.75 out / 1M |
| Gemini 3.8 Flash | Fast general-purpose assistant work | 1M tokens | No | $0.94 in · $4.69 out / 1M |
| Kimi K3 | Open-weight long-context reasoning | 1M tokens | Yes | $3.75 in · $18.75 out / 1M |
| Qwen 3.7 Plus | Balanced everyday chat and coding | 1M tokens | No | $0.50 in · $2 out / 1M |
MiniMax M3.1 Flash is the right pick when an agent has to actually do the work — long tool-call chains, repo-scale coding, computer-use steps — and the per-token bill matters across millions of steps. Choose Kimi K3 when open weights are non-negotiable, or Gemini 3.8 Flash when general-purpose polish matters more than unit economics.
Pricing
Billed per token on Venice: $0.19 per 1M input tokens and $0.75 per 1M output tokens.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Getting a key
Create a key in API settings. Nothing else is required to start.
Export it as VENICE_API_KEY so the snippets above run unedited.
Point an existing OpenAI client at https://api.venice.ai/api/v1. The scheme is part of the value: an OpenAI client given a bare host does not resolve it.
Pass minimax-m3-1-flash as the model and send the request.
FAQ
MiniMax M3.1 Flash is MiniMax's Flash-tier frontier language model, fully released on October 19, 2026 after previewing anonymously as "Space Bunny." It targets agentic reasoning, coding, computer use, and cybersecurity, with a 1,000,000-token context window, image and video input, web search, function calling, and a tunable thinking-depth setting.
On Venice, MiniMax M3.1 Flash is billed per token: $0.19 per 1M input tokens, $0.75 per 1M output tokens, and just $0.02 per 1M for cached input. There is no subscription required — you pay only for what you use.
It is not open source — Venice's model metadata lists the weights as closed, so it cannot be self-hosted or fine-tuned (though MiniMax's own documentation describes it as open-weight, so confirm licensing with MiniMax directly). It is not free to run, but Venice's per-token pricing at $0.19/$0.75 per 1M tokens makes it one of the cheapest frontier-tier options available.
Both offer 1M-token contexts, but they optimize for different things. MiniMax M3.1 Flash is tuned for agentic loops — coding, tool calls, computer use — and costs far less per token ($0.19/$0.75 vs $0.94/$4.69 per 1M on Venice). Gemini 3.8 Flash is the stronger pick for polished general-purpose assistant work where raw unit economics matter less.
Yes. Function calling, web search, and code-optimized reasoning are all listed capabilities, alongside vision (including multiple image inputs), video input, and reasoning. It is designed end-to-end for agent workflows that chain tools across many steps.
Venice runs it under the anonymized privacy tier: prompts are not stored, profiled, or used for training, and nothing ties your usage to a personal history. This is anonymized handling rather than end-to-end encryption or TEE isolation — Venice's strongest isolation tiers — but it still means zero retention of your conversations.
The effort parameter tunes how deeply the model reasons before answering, with five levels: low, medium, high, xhigh, and max. Higher levels think more thoroughly; when omitted, the default is max. Independent testing found that lowering effort to "low" dramatically cuts both latency and cost on routine tasks, so match the setting to the difficulty of each step.
MiniMax officially rolled out M3.1 Flash on October 19, 2026. An early version operated anonymously as "Space Bunny" starting in late September 2026, and a preview appeared on MiniMax Code in late September before the full public launch.
Yes. It accepts multimodal input including text, multiple images, and video, which makes it well suited to computer-use agents that read screenshots or screen recordings, document analysis with visual elements, and UI debugging workflows.
One key, free to start, no credit card.