MiniMax M3.1 Flash vs MiniMax M3
MiniMax M3.1 Flash (Oct 19) or the June flagship M3? M3.1 Flash lists at half M3's price with a quarter of the cache price, and MiniMax claims 70% lower agent-task cost and about 3× speed. Official specs, billing and published scores, with live CallAI pricing.
MiniMax M3.1 Flash
MiniMax M3
MiniMax hasn't compared the two on the same benchmarks: M3 was reported on Terminal-Bench 2.1 and OSWorld-Verified, M3.1 Flash on Terminal Bench 4.0 and OSWorld 2.0, so the scores don't line up. Price is clear: M3.1 Flash is half the unit price, a quarter of the cache price, has no 512K surcharge, and MiniMax claims 70% lower agent-task cost. Default batch and agent workloads to M3.1 Flash; for coding agents already tuned on M3, run a sample on both with the same key before switching.
Official specs and billing
| Benchmark | MiniMax M3.1 Flash | MiniMax M3 | Winner | Notes |
|---|---|---|---|---|
| List input / output price (per 1M tokens) | $0.15 / $0.60 | $0.30 / $1.20 | MiniMax M3.1 Flash | M3 at its "permanent 50% off" standard price, as of 2026-10-23 |
| Cache read (per 1M tokens) | $0.015 | $0.06 | MiniMax M3.1 Flash | Applies to repeated long system prompts |
| Above 512K input | No surcharge | $0.60 / $2.40 | MiniMax M3.1 Flash | M3 doubles the whole request above 512K input |
| Context window | 1M | 1M | Tie | Both take text, image and video input |
| Output speed | P50 170+ TPS | Not published | MiniMax M3.1 Flash | MiniMax says online inference is about 3× faster |
| Average agent-task cost | −70% | — | MiniMax M3.1 Flash | MiniMax figure, vs "the previous generation" |
MiniMax M3.1 Flash strengths & weaknesses
- Lists at $0.15 / $0.60 per 1M tokens, half of M3's standard price
- Cache reads at $0.015, a quarter of M3's
- MiniMax claims 70% lower total agent-task cost than the previous generation and 170+ TPS at P50
- No surcharge above 512K input; five reasoning levels from low to max
- All scores are self-reported; official Terminal Bench 4.0 is 38.1% vs 51.8% for Opus 5 in the same table
- The "70% lower cost" is against "the previous generation", which MiniMax does not name as M3
MiniMax M3 strengths & weaknesses
- MiniMax's June flagship: official SWE-Bench Pro 59.0%, Terminal-Bench 2.1 66.0%, OSWorld-Verified 70.06%
- Priority tier available at 1.5× price
- Out longer, so existing prompts and pipelines carry over
- Twice the unit price of M3.1 Flash, doubling again above 512K input
- Cache reads at $0.06, 4× M3.1 Flash
- MiniMax has not published a same-version benchmark comparison of the two
Scenario recommendations
| Scenario | Pick | Why |
|---|---|---|
| Bulk text processing, summaries, extraction | MiniMax M3.1 Flash | Half the unit price and a quarter of the cache price; the gap grows with volume. |
| Long-running agents (browser, computer use, tool calls) | MiniMax M3.1 Flash | Official OSWorld 2.0 61.7 and BankerToolBench 76.2, faster and cheaper per task. |
| Inputs above 512K tokens | MiniMax M3.1 Flash | M3 doubles its price above 512K; M3.1 Flash has no such tier. |
| Coding agents already tuned on M3 | Keep MiniMax M3 and A/B test | There are no same-version scores to compare and all figures are self-reported — verify on your own tasks first. |
Python dual-model gateway sample
OpenAI-compatibleOne Base URL and API key — switch models with the model parameter only:
import os
from openai import OpenAI
# One CallAI key — route freely between MiniMax M3.1 Flash and MiniMax M3
client = OpenAI(
api_key="sk-live-your-callai-key",
base_url="https://api.callaiapi.com/v1"
)
def query_model(model_id: str, prompt: str):
response = client.chat.completions.create(
model=model_id,
messages=[{"role": "user", "content": prompt}],
stream=True
)
for chunk in response:
print(chunk.choices[0].delta.content or "", end="", flush=True)
print("\n" + "="*40 + "\n")
# Model 1: MiniMax M3.1 Flash
print(">>> Output from: MiniMax M3.1 Flash")
query_model("minimax-m3.1-flash", "Briefly analyze the trade-offs of this architecture")
# Model 2: MiniMax M3
print(">>> Output from: MiniMax M3")
query_model("minimax-m3", "Briefly analyze the trade-offs of this architecture")Sources
FAQ
What do MiniMax M3.1 Flash and M3 cost on the API?
Official prices as of 2026-10-23: M3.1 Flash is $0.15 input, $0.60 output and $0.015 cache read per 1M tokens; M3 standard (after the permanent 50% discount) is $0.30 input, $1.20 output and $0.06 cache read, doubling above 512K input. The cards at the top show CallAI’s current price.
Why isn't there a benchmark table for both?
MiniMax used different benchmark versions for each (M3: Terminal-Bench 2.1, OSWorld-Verified; M3.1 Flash: Terminal Bench 4.0, OSWorld 2.0). Putting them side by side would mislead, so we only show figures with a primary source and matching methodology.
Can I trust M3.1 Flash's scores?
Every score on this page comes from MiniMax's blog and Hugging Face model card. They are vendor-reported and we have not verified them, so test on your own workload.
How do I call MiniMax M3.1 Flash?
Create an API key in the CallAI console, set the OpenAI SDK base_url to https://api.callaiapi.com/v1 and use model minimax-m3.1-flash. Pay as you go from $1, and the balance never expires.