MiniMax launched M3.1 Flash on October 19, 2026 at $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached input tokens on its official API. The price is flat: no peak hours, no long-context tier, no priority surcharge. It is half the rate of MiniMax M3, the model it builds on, and it is the exact off-peak rate of DeepSeek V4.1 Flash, charged around the clock instead of 133 hours a week.
The weights followed within the week under MiniMax's community license, and a third-party market formed faster than for any open-weight model we have tracked: eleven providers on OpenRouter by October 26, nine of them at exactly MiniMax's list price. This page has the official rates, how M3.1 Flash compares with M3 and with DeepSeek's Flash tier, every host, and what a real workload pays.
MiniMax M3.1 Flash official API pricing
Rates from MiniMax's platform pricing page, USD per 1M tokens:
| Token type | MiniMax M3.1 Flash | MiniMax M3, standard, up to 512K input | MiniMax M3, standard, above 512K input |
|---|---|---|---|
| Input | $0.15 | $0.30 | $0.60 |
| Cached input | $0.015 | $0.06 | $0.12 |
| Output | $0.60 | $1.20 | $2.40 |
| Pricing structure | Flat | 50 percent off a $0.60 / $2.40 list price | 50 percent off a $1.20 / $4.80 list price |
Two things about the structure matter more than the headline numbers.
M3.1 Flash is priced flat. M3 doubles its rate once a request passes 512K input tokens and charges 1.5x for a priority tier. M3.1 Flash has one price across its 1M token context window and no priority tier on the pricing page. For a long-context agent that routinely crosses 512K, the gap to M3 is not 2x but 4x on input.
M3's price is itself marked as a permanent 50 percent discount from a list price that no longer applies to anyone. The number buyers should compare against is the $0.30 and $1.20 they actually pay. M3.1 Flash halves that again. MiniMax's own framing is that overall usage cost on agentic tasks is 70 percent lower than its predecessor, which combines the lower rate with fewer tokens per task; that figure is MiniMax's and has not been independently reproduced.
What the model is. MiniMax describes M3.1 Flash as a multimodal mixture-of-experts model with about 428 billion total parameters and 23 billion active per token, built on M3's architecture with sparse attention throughout, quantization-aware training to 8-bit, and a new speculative decoding method. Context is up to 1M tokens. Reasoning effort is adjustable from low to max. MiniMax reports output above 200 tokens per second on its own platform; OpenRouter's measurement of the same endpoint is 69.
How M3.1 Flash compares with DeepSeek V4.1 Flash
The two models share a headline price and differ on the clock. USD per 1M tokens:
| Model | Input | Cached input | Output | When it applies |
|---|---|---|---|---|
| MiniMax M3.1 Flash | $0.15 | $0.015 | $0.60 | Always |
| DeepSeek V4.1 Flash, off-peak | $0.15 | $0.003 | $0.60 | 133 hours a week |
| DeepSeek V4.1 Flash, peak | $0.30 | $0.006 | $1.20 | 35 hours a week, Mon to Fri |
| MiniMax M3, standard, up to 512K | $0.30 | $0.06 | $1.20 | Always |
For an uncached workload that runs around the clock, M3.1 Flash is cheaper than DeepSeek V4.1 Flash on the official API, because DeepSeek bills 21 percent of a uniform week at double rate. For a cache-heavy workload, DeepSeek wins: its $0.003 cache-hit rate is a fifth of MiniMax's $0.015, and cache hits are where agent workloads spend most of their input budget. The schedule that decides DeepSeek's bill is in DeepSeek peak and off-peak hours.
Who hosts MiniMax M3.1 Flash and at what price
OpenRouter standard routing as of October 26, 2026. USD per 1M tokens, speed is median tokens per second.
| Host | Input | Output | Cached input | Speed |
|---|---|---|---|---|
| inference.net | $0.10 | $0.55 | $0.01 | 51 tps |
| GMICloud (5% promotion) | $0.1425 | $0.57 | $0.01425 | 65 tps |
| MiniMax official | $0.15 | $0.60 | $0.015 | 69 tps |
| Together, Parasail, Wafer, Nebius, ModelRun, DigitalOcean, Novita | $0.15 | $0.60 | $0.015 | 61 to 266 tps |
| SambaNova | $0.30 | $1.20 | $0.03 | 76 tps |
The pattern is unusual. Nine of eleven providers, MiniMax included, sit at exactly $0.15 and $0.60. One undercuts by a third on input, one charges double. When DeepSeek V4.1 Flash launched in September, the first third parties anchored on DeepSeek's peak rate, 2x the off-peak price, and took six weeks to move below it. MiniMax's hosts anchored on the one flat list price from the start, which says something about how low the official rate already is relative to serving cost.
Speed, not price, is what separates the hosts today. Parasail at 266 tokens per second and Wafer at 187 are nearly four times MiniMax's own OpenRouter endpoint at the same price. A buyer paying list has no reason to take the slowest route. Why token prices differ covers why a market this tight usually does not stay that way.
What a real workload pays
A production assistant doing 10 million input tokens and 1 million output tokens a day, uncached, at list rates on October 26, 2026.
| Route | Daily | Annual |
|---|---|---|
| M3.1 Flash, official or any list-price host | $2.10 | $767 |
| M3.1 Flash, inference.net | $1.55 | $566 |
| M3.1 Flash, SambaNova | $4.20 | $1,533 |
| MiniMax M3, official standard, under 512K | $4.20 | $1,533 |
| DeepSeek V4.1 Flash, official, all off-peak | $2.10 | $767 |
| DeepSeek V4.1 Flash, official, uniform traffic (21% peak) | $2.54 | $927 |
| DeepSeek V4.1 Flash, official, all peak | $4.20 | $1,533 |
Three things fall out of the table.
Against M3, the move is automatic. The same workload costs $4.20 a day on M3 standard and $2.10 on M3.1 Flash, before any gain from the shorter outputs MiniMax claims. A buyer on M3 who validates the new model halves the bill by changing the model name.
Against DeepSeek V4.1 Flash, the clock decides. A workload that can be scheduled entirely into DeepSeek's off-peak window ties MiniMax at $2.10. A workload that runs uniformly pays $2.54 on DeepSeek and $2.10 on MiniMax, a 17 percent gap. A workload that runs in business hours in Asia and Europe, which is where DeepSeek's peak window sits, pays $4.20 on DeepSeek and still $2.10 on MiniMax.
Cache flips it back. At a 60 percent cache hit rate, the M3.1 Flash official route drops to $1.29 a day, because 6 million input tokens bill at $0.015. The same workload on DeepSeek V4.1 Flash off-peak drops to $1.22, and DeepSeek's advantage grows with every point of cache hit rate above that. Cache-heavy agents should price both.
Which flash model and host to use
Running MiniMax M3: test M3.1 Flash and move. Half the price, and MiniMax says the quality is higher. There is no scenario on the pricing page where M3 is the better buy.
Uncached, round-the-clock, latency-sensitive: M3.1 Flash from a fast list-price host. Parasail, Wafer, and Together serve it at MiniMax's own price at two to four times the speed.
Cost-first and throughput-tolerant: inference.net at $0.10 and $0.55 is the floor today, at 51 tokens per second.
Cache-heavy agents: price DeepSeek V4.1 Flash against it. The $0.003 cache rate is the lowest in this class and no MiniMax host matches it.
Already on DeepSeek V4.1 Flash with peak-hour traffic: M3.1 Flash is the first model at DeepSeek's off-peak rate that does not care what time it is. That alone is worth an eval.
The model-by-model view across DeepSeek, GLM, Qwen, Kimi, and now MiniMax is in the LLM API pricing comparison. To track these rates as hosts move, see the Mercatus Token Index.
What this launch says about flash-tier pricing
In the past seven weeks, two vendors have shipped a new small-model tier at $0.15 input and $0.60 output: DeepSeek on September 10 and MiniMax on October 19. Neither cut the price of an existing product to get there. Both released a better model at a lower rate and left the old one on the price list. Better models create competition, and in the flash tier the competition has now produced a common price from two sellers on two continents, with a third-party market that matched it inside a week.
What buyers still do not have is a way to hold that price. MiniMax's pricing page carries no end date and no promotional flag, but it also carries no commitment, and M3's own "permanent 50 percent off" label shows how quickly a list price can be redefined. For a team moving real volume onto M3.1 Flash, how to lock in token prices covers what a forward commitment does, and how a token exchange works covers how a single market price forms once eleven sellers quote the same model.
Frequently asked questions
How much does the MiniMax M3.1 Flash API cost?
$0.15 per million input tokens, $0.015 per million cached input tokens, and $0.60 per million output tokens on MiniMax's official API. The price is flat at all hours and across the full context window.
Is MiniMax M3.1 Flash cheaper than MiniMax M3?
Yes, by half. M3 standard is $0.30 input and $1.20 output for requests up to 512K input tokens, and $0.60 and $2.40 above that. M3.1 Flash is $0.15 and $0.60 at any length.
Is MiniMax M3.1 Flash cheaper than DeepSeek V4.1 Flash?
Same headline price, different clock. Both are $0.15 input and $0.60 output, but DeepSeek doubles that during 35 peak hours a week and MiniMax does not. DeepSeek's cache-hit rate of $0.003 is lower than MiniMax's $0.015, so cache-heavy workloads can still be cheaper on DeepSeek.
Who hosts MiniMax M3.1 Flash?
Eleven providers on OpenRouter as of October 26, 2026: MiniMax, inference.net, GMICloud, DigitalOcean, NovitaAI, Nebius, Wafer, ModelRun, Parasail, Together, and SambaNova. Nine list at exactly $0.15 and $0.60. inference.net is lowest at $0.10 and $0.55.
Is MiniMax M3.1 Flash open weight?
Yes. The weights are on Hugging Face under the MiniMax community license, in 8-bit MXFP8 format, with about 428 billion total and 23 billion active parameters.
What is the context window of MiniMax M3.1 Flash?
Up to 1,000,000 tokens, priced flat with no long-context tier.
Methodology
M3.1 Flash and M3 rates, cache rates, the 512K pricing threshold, and priority tier terms are from MiniMax's platform pricing page. Launch date, speed claims, and the 70 percent usage-cost claim are from MiniMax's M3.1 Flash announcement of October 19, 2026. Parameter counts, context length, license, and architecture notes are from the model's Hugging Face page. Third-party host prices, cache rates, and throughput are from OpenRouter's model page, standard routing, on October 26, 2026, with the GMICloud promotional price noted. DeepSeek V4.1 Flash rates and peak-hour definitions are from DeepSeek's API documentation as documented on the Mercatus DeepSeek V4.1 Flash page. Workload calculations assume uncached input except where a cache rate is stated; the uniform-traffic case weights peak hours at 35 of 168 weekly hours. Last verified: 2026-10-26.