MiniMax M3 API Pricing and Speed
MiniMax M3 API pricing: $0.30 per 1M tokens for input and $1.20 per 1M tokens for output, from the official API price list of MiniMax. Official rate checked (source); OpenRouter is a separate channel below. Our measured latency and output speed are shown by route and country below. Published OpenRouter: Last checked · Source (data) · Methodology
OpenRouter: Amount worked out from the source page: $0.30 per 1M tokens, in USD · as shown in the United States on 9 Nov 2026 · the page does not say whether tax is included; we do not assume it is excluded · no price change seen since 16 Oct 2026.
Quality on your task: not measured by Measured Lab yet. On our data, MiniMax M3 fits 4 of 6 common tasks by published prices, context window and terms, with calculated monthly costs.
What can I use it for?
Six common tasks cost a month at list prices
- Calculated Answer customer messages: $5.85 for 5,000 customer messages a month
- Unknown Pull data from invoices and documents with strict JSON and page images: not confirmed, strict JSON output not in our data
- Calculated Summarize documents and meetings: $2.79 for 1,000 documents or meetings a month
- Calculated Translate content: $4.14 for 3,000 pages a month
- Calculated Chat assistant on your website: $59.85 for 10,000 conversations a month
- Unknown Write and review code with tool calls: not confirmed, tool calls not in our data
What our data confirms
- Published Reads images; OpenRouter listing, 9 Nov 2026
- Published Weights on Hugging Face: MiniMaxAI/MiniMax-M3, as linked by the OpenRouter listing
Example to try
“Draft a reply to this refund request from our returns policy, and hand it to a person if the order is over $500.” Run it on your own data before you choose.
What we measured
- Observed Speed on a short test request: first token 3.9 s, 463 tokens/s (median of 88 test requests on 23 dates, short request)
- Observed Hidden reasoning: 506 tokens on that request, billed as output
Stated by the provider
- Stated by the provider MiniMax for this model: SWE-Bench Pro 59.0%, Terminal-Bench 2.1 66.0%, MCP Atlas 74.2%; quotes and test conditions in Quality signals
Before you choose
- Unknown Training on your data: terms not in our data
- Unknown Where requests are processed: Unknown
Prices USD per 1M tokens, as listed
- $0.30Input per 1M tokens, from MiniMax Published
- $1.20Output per 1M tokens, from MiniMax Published
All prices by channel
| Price per 1M tokens | MiniMax direct | Through OpenRouter | Previous prices by channel |
|---|---|---|---|
| Input | $0.30 Published | $0.30 Published | — |
| Output | $1.20 Published | $1.20 Published | — |
| Cached input · read | UnknownUnknown | $0.06 Published | — |
| Batch · input / output | — | $0.224 / $0.88 Published | — |
Input price from MiniMax, 24 Oct 2026 – 9 Nov 2026
24 Oct 2026$0.30 throughout9 Nov 2026
Input price through OpenRouter, 16 Oct 2026 – 9 Nov 2026
16 Oct 2026$0.30 throughout9 Nov 2026
Context window1,048,576tokens Published
Max output512,000tokens Published
OpenRouter providers who serves this model through OpenRouter, at what price
OpenRouter sends a call without a pinned provider to one of these endpoints, and each endpoint has its own price. Prices are USD per 1M tokens, input / output.
Pinned min: $0.224 / $0.88 · DeepInfra (deepinfra/fp8/flex)
Unpinned range: input $0.224–$0.45, output $0.88–$1.80 across 12 endpoints of 10 providers (listing shows $0.30 / $1.20).
Models with the widest provider price spread on OpenRouter
Request fragment that pins DeepInfra (deepinfra/fp8/flex):
{"provider": {"order": ["deepinfra/fp8/flex"], "allow_fallbacks": false}}With fallbacks turned off, OpenRouter does not move the call to another provider. A full endpoint tag pins that endpoint; the provider name alone covers all of its endpoints (OpenRouter provider routing).
All endpoints (12), by input price
| Provider | Endpoint tag | Precision | Input | Output | Context, tokens |
|---|---|---|---|---|---|
| DeepInfra | deepinfra/fp8/flex | fp8 | $0.224 | $0.88 | 524,288 |
| CoreWeave | coreweave/fp4 | fp4 | $0.23 | $0.96 | 262,144 |
| GMICloud | gmicloud/fp8 | fp8 | $0.24 | $0.96 | 1,048,576 |
| DeepInfra | deepinfra/fp8 | fp8 | $0.28 | $1.10 | 524,288 |
| AtlasCloud | atlas-cloud/fp8 | fp8 | $0.30 | $1.20 | 524,300 |
| Minimax | minimax | Not stated | $0.30 | $1.20 | 524,288 |
| Novita | novita/fp8 | fp8 | $0.30 | $1.20 | 1,000,000 |
| Parasail | parasail/fp8 | fp8 | $0.30 | $1.20 | 1,048,576 |
| StreamLake | streamlake/fp8 | fp8 | $0.30 | $1.20 | 1,000,000 |
| Venice | venice/fp8 | fp8 | $0.30 | $1.20 | 524,288 |
| ModelRun | modelrun/fp8 | fp8 | $0.30 | $1.30 | 1,048,576 |
| Parasail | parasail/fast | fp8 | $0.45 | $1.80 | 1,048,576 |
Sorted by input price, then output price. Precision is the quantization field of the endpoints list.
Measured by us our own short test request: about 380 tokens in, a short JSON answer out · medians from answers on two dates or more
United States
| Route | First answer token | Output speed | Cost per request | Hidden reasoning |
|---|---|---|---|---|
| Direct API · MiniMaxreasoning: low88 answers on 23 dates · 76 % answered | 3.86 srange 1.76–10.61 s · slow 10 %: 5.90 s Observed | 463 tokens/srange 193–711 · slowest 10 %: 264 tokens/s | $0.000855range $0.00052–$0.00139 Calculated | 506for 69 visible |
United Kingdom
| Route | First answer token | Output speed | Cost per request | Hidden reasoning |
|---|---|---|---|---|
| Direct API · MiniMaxreasoning: low79 answers on 19 dates · 75 % answered | 4.08 srange 1.30–10.50 s · slow 10 %: 6.76 s Observed | 470 tokens/srange 201–659 · slowest 10 %: 288 tokens/s | $0.00083range $0.000321–$0.00139 Calculated | 486for 69 visible |
Germany
| Route | First answer token | Output speed | Cost per request | Hidden reasoning |
|---|---|---|---|---|
| Direct API · MiniMaxreasoning: low82 answers on 21 dates · 78 % answered | 3.28 srange 1.97–8.78 s · slow 10 %: 6.57 s Observed | 386 tokens/srange 98.6–703 · slowest 10 %: 253 tokens/s | $0.000794range $0.000397–$0.00139 Calculated | 459for 68 visible |
Canada
| Route | First answer token | Output speed | Cost per request | Hidden reasoning |
|---|---|---|---|---|
| Direct API · MiniMaxreasoning: low75 answers on 19 dates · 71 % answered | 3.62 srange 1.97–8.28 s · slow 10 %: 6.02 s Observed | 428 tokens/srange 237–746 · slowest 10 %: 271 tokens/s | $0.000806range $0.000526–$0.00139 Calculated | 464for 69 visible |
Australia
| Route | First answer token | Output speed | Cost per request | Hidden reasoning |
|---|---|---|---|---|
| Direct API · MiniMaxreasoning: low84 answers on 20 dates · 80 % answered | 4.27 srange 2.28–8.84 s · slow 10 %: 6.21 s Observed | 386 tokens/srange 45.4–689 · slowest 10 %: 281 tokens/s | $0.000795range $0.000562–$0.00139 Calculated | 463for 68 visible |
- Hidden reasoning tokens are billed as output, at the output price, even though they are not part of the answer.
How the test works and what each number means: methodology. The figures are also in our open data.
Quality signals public test results, each with its source and date
| Signal | Result | Source |
|---|---|---|
| Instruction followingIFEval, 50 prompts · reasoning: low | 92 % Observedof 50 prompts with every instruction followed | Our run of a fixed IFEval subset, checked by code92 % on 22 Oct 2026 · 90 % on 17 Oct 2026 · instruction level 94 % |
Figures the provider publishes (6)
| Benchmark | Provider's figure | Where the provider says it |
|---|---|---|
| SWE-Bench Pro | 59.0% Stated by the provider | SWE-Bench Pro: 59.0%MiniMax M3: Frontier Coding, 1M Context, Native Multimodality — All in One Model - MiniMax Researchread 17 Oct 2026 |
| Terminal-Bench 2.1 | 66.0% Stated by the provider | Terminal-Bench 2.1: 66.0%MiniMax M3: Frontier Coding, 1M Context, Native Multimodality — All in One Model - MiniMax Researchread 17 Oct 2026 |
| BrowseComp | 83.5 Stated by the provider | On BrowseComp, M3 scores 83.5, surpassing Opus 4.7 (79.3), demonstrating strong autonomous browsing and information retrieval capabilities.MiniMax M3 - Coding & Agentic Frontier, 1M Context, Multimodal | MiniMaxread 17 Oct 2026 |
| MCP Atlas | 74.2% Stated by the provider | MCP Atlas: 74.2%MiniMax M3: Frontier Coding, 1M Context, Native Multimodality — All in One Model - MiniMax Researchread 17 Oct 2026 |
| SWE-fficiency | 34.8% Stated by the provider | SWE-fficiency: 34.8%MiniMax M3: Frontier Coding, 1M Context, Native Multimodality — All in One Model - MiniMax Researchread 17 Oct 2026 |
| OSWorld-Verified | 70.06% Stated by the provider | Increasing Max Steps from 100 → 200 improved task completion rate from 68.70% → 70.06%.MiniMax M3: Frontier Coding, 1M Context, Native Multimodality — All in One Model - MiniMax Researchread 17 Oct 2026 |
- Your own task: quality on it is not measured here. These are public test sets; a short trial on a few of your own examples is the check that fits your work.
- Other public leaderboards, without figures copied here: human preference votes on open-ended answers: Arena (LMArena) text leaderboard; long tasks in coding, data analysis and research: LiveBench.
What each signal covers and how our run is scored: methodology.
How we got this number
- What we read: the published price on openrouter.ai/api/v1/models
- Checked: ; tracked since 16 Oct 2026
- Shown in: the United States
- Checks so far: 25 checks since 16 Oct 2026
- What changed: no price change since 16 Oct 2026
- Page updated:
Other measured facts
| Question | Answer | Checked | Status |
|---|---|---|---|
| How much does MiniMax M3 cost per 1M cache-read tokens (cache hits) on OpenRouter? | $0.06 per 1M tokens | Published | |
| How much does MiniMax M3 cost per 1M output tokens on OpenRouter? | $1.20 per 1M tokens | Published | |
| How much does MiniMax-M3 cost per 1M input tokens directly from MiniMax (official API price)? | $0.30 per 1M tokens | Published | |
| How much does MiniMax-M3 cost per 1M output tokens directly from MiniMax (official API price)? | $1.20 per 1M tokens | Published | |
| What is the context window of MiniMax M3 on OpenRouter, in tokens? | 1,048,576 tokens | Published | |
| What is the maximum output length of MiniMax M3 on OpenRouter, in tokens? | 512,000 tokens | Published |
History
Same price in all 25 checks since 16 Oct 2026.
- — First check: $0.30 per 1M tokens
The source record lists $0.0000003 per token ($0.30 per 1M tokens), fetched on 9 Nov 2026.
The source record, as published
"prompt":"0.0000003"
Latest briefings each price change recalculated as a task cost
- DeepSeek V4 Pro 0813: price rise input $0.115 → $0.111 per 1M tokens (-3 %), output $4.00 → $4.44 per 1M tokens (+11 %)
- DeepSeek V4.1 Flash 1017: price cut input $0.300 → $0.0060 per 1M tokens (-98 %), output $1.20 → $0.500 per 1M tokens (-58 %)
- MoonshotAI Kimi K2.5: price rise input $0.500 → $0.532 per 1M tokens (+6 %), output $2.50 → $3.33 per 1M tokens (+33 %)
Record history — 147 records for 7 questions
- Last record
- Shown in
- the United States
| Time | Value | Status |
|---|---|---|
| 9 Nov 2026 | $0.30 per 1M tokens | Published |
| 8 Nov 2026 | $0.30 per 1M tokens | Published |
| 7 Nov 2026 | $0.30 per 1M tokens | Published |
| 6 Nov 2026 | $0.30 per 1M tokens | Published |
| 5 Nov 2026 | $0.30 per 1M tokens | Published |
| 4 Nov 2026 | $0.30 per 1M tokens | Published |
| 3 Nov 2026 | $0.30 per 1M tokens | Published |
| 2 Nov 2026 | $0.30 per 1M tokens | Published |
| 1 Nov 2026 | $0.30 per 1M tokens | Published |
| Time | Value | Status |
|---|---|---|
| 9 Nov 2026 | $0.06 per 1M tokens | Published |
| 8 Nov 2026 | $0.06 per 1M tokens | Published |
| 7 Nov 2026 | $0.06 per 1M tokens | Published |
| 6 Nov 2026 | $0.06 per 1M tokens | Published |
| 5 Nov 2026 | $0.06 per 1M tokens | Published |
| 4 Nov 2026 | $0.06 per 1M tokens | Published |
| 3 Nov 2026 | $0.06 per 1M tokens | Published |
| 2 Nov 2026 | $0.06 per 1M tokens | Published |
| 1 Nov 2026 | $0.06 per 1M tokens | Published |
| Time | Value | Status |
|---|---|---|
| 9 Nov 2026 | $1.20 per 1M tokens | Published |
| 8 Nov 2026 | $1.20 per 1M tokens | Published |
| 7 Nov 2026 | $1.20 per 1M tokens | Published |
| 6 Nov 2026 | $1.20 per 1M tokens | Published |
| 5 Nov 2026 | $1.20 per 1M tokens | Published |
| 4 Nov 2026 | $1.20 per 1M tokens | Published |
| 3 Nov 2026 | $1.20 per 1M tokens | Published |
| 2 Nov 2026 | $1.20 per 1M tokens | Published |
| Time | Value | Status |
|---|---|---|
| 9 Nov 2026 | $0.30 per 1M tokens | Published |
| 8 Nov 2026 | $0.30 per 1M tokens | Published |
| 7 Nov 2026 | $0.30 per 1M tokens | Published |
| 6 Nov 2026 | $0.30 per 1M tokens | Published |
| 5 Nov 2026 | $0.30 per 1M tokens | Published |
| 4 Nov 2026 | $0.30 per 1M tokens | Published |
| 3 Nov 2026 | $0.30 per 1M tokens | Published |
| 2 Nov 2026 | $0.30 per 1M tokens | Published |
| Time | Value | Status |
|---|---|---|
| 9 Nov 2026 | $1.20 per 1M tokens | Published |
| 8 Nov 2026 | $1.20 per 1M tokens | Published |
| 7 Nov 2026 | $1.20 per 1M tokens | Published |
| 6 Nov 2026 | $1.20 per 1M tokens | Published |
| 5 Nov 2026 | $1.20 per 1M tokens | Published |
| 4 Nov 2026 | $1.20 per 1M tokens | Published |
| 3 Nov 2026 | $1.20 per 1M tokens | Published |
| 2 Nov 2026 | $1.20 per 1M tokens | Published |
| 1 Nov 2026 | $1.20 per 1M tokens | Published |
| Time | Value | Status |
|---|---|---|
| 9 Nov 2026 | 1,048,576 tokens | Published |
| 8 Nov 2026 | 1,048,576 tokens | Published |
| 7 Nov 2026 | 1,048,576 tokens | Published |
| 6 Nov 2026 | 1,048,576 tokens | Published |
| 5 Nov 2026 | 1,048,576 tokens | Published |
| 4 Nov 2026 | 1,048,576 tokens | Published |
| 3 Nov 2026 | 1,048,576 tokens | Published |
| 2 Nov 2026 | 1,048,576 tokens | Published |
| 1 Nov 2026 | 1,048,576 tokens | Published |
| Time | Value | Status |
|---|---|---|
| 9 Nov 2026 | 512,000 tokens | Published |
| 8 Nov 2026 | 512,000 tokens | Published |
| 7 Nov 2026 | 512,000 tokens | Published |
| 6 Nov 2026 | 512,000 tokens | Published |
| 5 Nov 2026 | 512,000 tokens | Published |
| 4 Nov 2026 | 512,000 tokens | Published |
| 3 Nov 2026 | 512,000 tokens | Published |
| 2 Nov 2026 | 512,000 tokens | Published |
Showing the latest 60 of 147 records; the full history is in data.json and data.csv.