FLAT RATE TO 1M1M CONTEXTIMAGE + VIDEO INTHINKING EFFORTWEIGHTS PROMISED
MiniMax M3.1 Flash API Pricing
MiniMax M3.1 Flash is MiniMax's ultra-efficient model for agents, coding and computer use, released on October 19, 2026. It costs $0.15/M input, $0.015/M cached input and $0.60/M output, one flat card across its 1M-token window, where MiniMax M3 doubles its rates above 512K and sells a Priority tier on top. It takes text, images and video, thinks before it answers, and with its launch MiniMax took the whole M2.x family off its price list.
Input - per 1M tokens
$0.15/M
MiniMax M3 $0.30/M up to 512K flat to 1M
Output - per 1M tokens
$0.60/M
MiniMax M3 $1.20/M up to 512K thinking billed here
Cached input - prompt cache read
$0.015/M
Cache read one rate to 1M -90%
Effective - agentic blend
$0.0842/M
92/8 split - 82% cache
§ 01 / TERMINAL
MiniMax M3.1 Flash API cost calculator.
Calculator pre-loaded with MiniMax M3.1 Flash's flat rates. Thinking is on by default at the highest effort, and thinking tokens bill as output, so a lower effort setting is the main lever on the output share.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
MiniMax M3.1 Flash cost on real workloads.
CODING AGENT
Repo repair task
$0.00476/task
COMPUTER USE
Desktop workflow step
$0.00139/step
LONG CONTEXT
Read a whole repository
$0.028/pass
OFFICE
Analyst memo draft
$0.00337/memo
Calibrated · measured on the vendor's tokenizer · 2026-06-10 Auto-counts as you type
Counts use a chars-per-token calibration measured on the vendor's own published tokenizer (MiniMaxAI/MiniMax-M2.1, 2026-06-10). English prose is typically within a few percent; code and non-Latin scripts tokenize heavier. For billing-exact counts use the vendor's count-tokens API.
Characters 402
Words 65
Tokens (estimated) 75 tokens
Cost as input · uncached $0.00001 USD
Cost as output · uncached $0.00005 USD
Cost as cached input $0.000001 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| MiniMax M3.1 Flash Current | $0.15 cache $0.015 | $0.60 | $0.0842 agentic 92/8 | 1M | Cheap 1M-context agents, coding and computer use |
| MiniMax M3 | $0.30 cache $0.06 | $1.20 | $0.191 +2.3× | 1M | MiniMax's frontier model, with a Priority tier |
| MiniMax M2.7 | $0.30 cache $0.06 | $1.20 | $0.191 deprecated · +2.3× | 205K | The previous generation, off the price list |
| DeepSeek V4.1 Flash | $0.30 cache $0.006 | $1.20 | $0.15 +78% | 1M | Cheap 1M-context agent loops with images |
| GLM-5.3-Flash | $0.15 cache $0.03 | $0.50 | $0.0875 +4% | 1M | Z.ai's multimodal flash tier |
| MiMo-V2.6-Flash | $0.14 cache $0.0028 | $0.28 | $0.0477 −43% | 1M | Omni-modal input with a Batch API |
| Mercury 3 | $0.20 cache $0.02 | $0.70 | $0.104 +24% · promo | 1M | Diffusion LLM sold on speed |
| Gemini 3.8 Flash | $0.75 cache $0.075 | $3.75 | $0.481 +5.7× · promo | 1M | The Western flash tier |
MiniMax M3.1 Flash has published scores on 10 benchmarks we track, and leads 3 of them. All of them are the lab's own reports; each row links to the page that published it.
| Benchmark | Score | Rank | Leader | Source |
|---|---|---|---|---|
| AutomationBench Agents and computer use · Main set (as reported) | 62.8% v1.0.6, 600 public tasks | #1 of 22 | MiniMax M3.1 Flash leads | Vendor-reported MiniMax |
| DeepSWE Coding | 73.7% as reported | #6 of 27 | Gemini 4 Argon · 77.9% | Vendor-reported MiniMax |
| NL2Repo-Bench Coding | 61.2% as reported | #3 of 8 | DeepSeek V4.1 Flash · 68.2% | Vendor-reported MiniMax |
| Terminal-Bench 4.0 Agents and computer use | 38.1% as reported | #15 of 34 | Kimi K3.1 Pro · 65.7% | Vendor-reported MiniMax |
| Agents' Last Exam Agents and computer use · Split not stated | 29.6% split not stated | #2 of 4 | MiMo-V2.6-Pro · 31.6% | Vendor-reported MiniMax |
| OSWorld 2 Agents and computer use · 2.0, scoring not stated | 61.7% scoring not stated | #1 of 2 | MiniMax M3.1 Flash leads | Vendor-reported MiniMax |
| PostTrainBench AI research | 39.3% v1.1 | #3 of 6 | Gemini 4 Argon · 45.3% | Vendor-reported MiniMax |
| Humanity's Last Exam Expert knowledge · No tools | 43.4% no tools | #7 of 11 | Claude Sonnet 5.5 · 56.9% | Vendor-reported MiniMax |
| CyberGym Cybersecurity | 82.2% as reported | #6 of 8 | MiMo-V2.6-Flash · 95.1% | Vendor-reported MiniMax |
| CharXiv Reasoning Vision and documents · Tool use not stated | 82.5% tool use not stated | #1 of 1 | MiniMax M3.1 Flash leads | Vendor-reported MiniMax |
MiniMax M3.1 Flash API pricing FAQ.
What MiniMax M3.1 Flash costs, how it differs from MiniMax M3, and what its launch did to the M2.x models.
Q · 01 How much does MiniMax M3.1 Flash cost? +
$0.15/M input, $0.015/M cached input and $0.60/M output, the same on MiniMax's pricing page and in the launch post. That is $0.08/M on our 92/8 agentic blend with 82% cache hits. MiniMax prints one row for it: no long-context step and no Priority tier.Q · 02 How does it compare with MiniMax M3? +
It is the cheaper, faster model in the same family. MiniMax M3 lists
$0.30 input and $1.20 output up to 512K tokens, a higher rate above 512K and a Priority tier on top; M3.1 Flash has one flat rate across its whole 1M window. Both take text, image and video input. MiniMax keeps M3 available and says a Pro model will follow.Q · 03 What happened to MiniMax M2.7 and the rest of M2? +
They are off MiniMax's price list. Around the M3.1 Flash launch on October 19, 2026, MiniMax dropped M2.7, M2.5, M2.1, M2 and their Highspeed variants from its pricing page and from its list of supported models, without a deprecation notice or a shutdown date. Their pages here are kept with the last published rates; M2.7 listed at
$0.30 input and $1.20 output.Q · 04 Does thinking cost extra? +
Not as a separate rate, but it adds output tokens. Thinking is on by default, and the
effort setting runs from low to max; when you omit it, M3.1 Flash uses max. Higher effort thinks longer and produces more output tokens, so on short tasks it is worth setting effort explicitly.Q · 05 Is it open weight? +
Not yet. The launch post says the weights "will be released later". MiniMax already publishes open weights for M3 and M2.7, which its self-hosting guide covers.
Q · 06 How fast is it? +
MiniMax says output reaches 200+ tokens per second in everyday use, with a median above 170, and that agentic usage costs 70% less than its predecessor; it also reports 73.7% on DeepSWE v1.1 and 61.7 on OSWorld V2. These are MiniMax's figures, not our measurements.