MiniMax · open weights
MiniMax M2.7 API
The middle of the catalogue — 10B active parameters, half the price of Kimi, and a SWE-Bench Pro score matching GPT-5.3-Codex.
~40% below OpenRouter list for the same model. Prices verified August 14, 2026. Full market table on /pricing.
Call it
Model id: MiniMaxAI/MiniMax-M2.7 (case-sensitive). Any OpenAI SDK works.
from openai import OpenAI
client = OpenAI(
base_url="https://router.mingles.ai/v1",
api_key="sk-your-free-key",
)
resp = client.chat.completions.create(
model="MiniMaxAI/MiniMax-M2.7",
messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)Published benchmarks
| Benchmark | Score | As published |
|---|---|---|
| SWE-Bench Pro | 56.2% | matches GPT-5.3-Codex |
| Terminal-Bench 2 | 57.0% | — |
| SWE Multilingual | 76.5 | — |
| VIBE-Pro | 55.6% | near Claude Opus 4.6 |
| MLE-Bench Lite | 66.6% | medal rate; behind Opus 4.6 and GPT-5.4 |
| Toolathon | 46.3% | — |
Source: MiniMax — MiniMax-M2.7 model card. Full comparison with sources on /compare/claude-api.
Specs
| Architecture | Mixture-of-experts, 229B total / 10B active parameters (per model card) |
| Strengths | High-volume agent loops, SWE-Bench Pro 56.2% (matches GPT-5.3-Codex per card) |
| License | Open weights (see model card) |
On a plan
Plans hold a priced allowance, not a token count, so what it buys depends on the model. On MiniMax M2.7 it works out to roughly:
M2.7 is the middle option: only 10B active parameters, so it is priced at half of Kimi per blended token while matching GPT-5.3-Codex on SWE-Bench Pro by its own card. Since DeepSeek V4 Flash arrived at roughly a third of M2.7's blended price, the honest use for this model is the middle of the range: reach for it when DeepSeek is not holding up on a task and Kimi is more than you need. Note the two cards are not comparable — MiniMax and DeepSeek report different benchmarks under their own harnesses, so this is a call your own evals have to make.
FAQ
▸How do I call MiniMax M2.7?
POST /v1/chat/completions with "model": "MiniMaxAI/MiniMax-M2.7" (case-sensitive) against https://router.mingles.ai/v1 — any OpenAI SDK or agent tool works.
▸What does MiniMax M2.7 cost?
$0.18 per 1M input and $0.72 per 1M output. Since 2026-08-14 it is no longer the cheapest model we serve: DeepSeek V4 Flash is about a third of the price, and the "up to N tokens" plan figures are now quoted in that instead.
▸MiniMax, DeepSeek or Kimi?
The burn rates shown in the console: DeepSeek V4 Flash ×1, MiniMax-M2.7 ×2.9, Kimi-K2.6 ×5.7. Cheap first pass on DeepSeek, MiniMax when it needs more, Kimi where quality clearly shows — most workloads end up mixed.