MiniMax M3 API — Hosted, Private, Long-Context Inference
Call MiniMax M3 through an OpenAI-compatible API on attested, verifiable infrastructure. ZDR and TEE tiers, automatic fallback, per-model pricing.
MiniMax M3 on attested hardware, with the privacy tier chosen per request.
M3 is the June 2026 open-weights release that agentic-coding and long-context teams adopted fast. It is a 428B multimodal model. Launch coverage benchmarked it near frontier quality at a fraction of frontier prices, and within weeks it was top-5 by token share on OpenRouter. Self-hosting 428B parameters is impractical for most teams, and only a handful of providers host it.
TrustedRouter serves minimax/minimax-m3 through an OpenAI-compatible API. Point your SDK at a new base_url and keep your code. MiniMax's own API is one of the M3 routes, alongside third-party hosts with Zero-Data-Retention routes. With Confidential privacy (provider.min_privacy "confidential"), prompts never reach the model vendor: a Confidential request routes only to third-party confidential-compute hosts, and fails rather than fall back to the vendor's own API. One licensing note: M3 ships under the MiniMax Community License, open weights but not MIT or Apache.
from openai import OpenAI
client = OpenAI(
base_url="https://api-azure.trustedrouter.com/v1",
api_key=TRUSTEDROUTER_API_KEY,
)
resp = client.chat.completions.create(
model="minimax/minimax-m3",
messages=[{"role": "user",
"content": "Plan the refactor of this repo."}],
)
print(resp.choices[0].message.content)ZDR routes
Pick Zero-Data-Retention routes, where providers contractually keep nothing. TrustedRouter never logs prompt or output content. Ordinary synchronous and streaming inference does not retain it. The opt-in Batch API temporarily retains enclave-encrypted artifacts for up to 30 days. Details at /security.
Fallback across M3 providers
Few providers host MiniMax M3. TrustedRouter routes across them and rolls over automatically when one goes down. Latency is measured from real routed requests and published at /llm-provider-latency-benchmarks.
Per-model pricing, thin markup
Every M3 route is priced per model at /pricing, a thin markup over provider list prices. Pay with prepaid credits, bring your own key, or pay per request with x402. MetaMask sign-in works without a traditional account.
Attestation beats a privacy policy.
Every M3 prompt runs inside a hardware-attested confidential-compute gateway. The gateway source is open and the image digest is published. Ask it to prove itself: curl "https://api-azure.trustedrouter.com/attestation?nonce=$NONCE" returns a JWT signed by the CPU vendor's root key, bound to your live TLS session so it cannot be replayed. See /security.
Routing path matters for Chinese open weights. In our GLM test, the same weights answered 60 of 60 sensitive questions through the enclave versus 30 of 60 via the vendor-hosted API. The filtering is host-applied, not in the weights. One proof point on scale: a document-processing company scaled from zero to over a billion tokens per week through the attested gateway within three weeks of onboarding. Also on the roster: Kimi K2.7, among 220+ model routes across 30+ providers.
Current routes, prices, privacy, and measured performance.
Catalog facts come from the routes currently configured in TrustedRouter. Performance uses the same cached metadata snapshot as the public leaderboard. Prompts and outputs are not part of these measurements.
| Model | Providers | Context | Input | Output | Privacy | Measured route |
|---|---|---|---|---|---|---|
MiniMax: MiniMax M3minimax/minimax-m3 |
+13
|
1,048,576 | $0.24265/1M to $0.633/1M | $1.0128/1M to $2.532/1M | ZDR 4 cited scores | No measurements yet |
MiniMax: MiniMax M2.5minimax/minimax-m2.5 |
+2
|
196,608 | $0.311225/1M to $0.37347/1M | $1.266/1M to $1.49388/1M | varies | No measurements yet |
MiniMax: MiniMax M2.7minimax/minimax-m2.7 |
+1
|
196,608 | $0.22155/1M to $0.3165/1M | $1.266/1M | ZDR | No measurements yet |
MiniMax-M3.1-Flashminimax/minimax-m3.1-flash |
|
1,000,000 | $0.15825/1M | $0.633/1M | varies | No measurements yet |
MiniMax-M3MiniMaxAI/MiniMax-M3 |
|
1,048,576 | $0.3165/1M | $1.266/1M | ZDR | No measurements yet |
MiniMax M2.5minimax/minimax-m25 |
|
198,000 | $0.28485/1M | $1.00225/1M | varies | No measurements yet |
Browse every modelReview provider policiesOpen the full leaderboardSnapshot 2026-10-22T19:44:43Z
Questions
Is MiniMax M3 open source?
M3 is an open-weights release under the MiniMax Community License. That license is not MIT or Apache, so review its terms before building on the weights directly. Most teams do not self-host a 428B-parameter multimodal model anyway. Through TrustedRouter you call hosted M3 routes over an OpenAI-compatible API, with Zero-Data-Retention and TEE privacy tiers available per route.
What does the MiniMax M3 API cost through TrustedRouter?
Pricing is per model and published on the pricing page, with a thin markup over provider list prices. Prices drift, so we do not quote numbers on marketing pages. You can pay with prepaid credits via Stripe, bring your own provider key, or pay per request with x402. MetaMask wallet sign-in works without a traditional account.
Do my prompts go to MiniMax when I use this API?
They can on Standard routing, where MiniMax's own API is one of the M3 routes. With Confidential privacy (provider.min_privacy "confidential"), prompts never reach the model vendor: a Confidential request routes only to third-party confidential-compute hosts, and fails rather than fall back to the vendor's own API. Zero-Data-Retention routes are available. TrustedRouter never logs prompt or output content. Ordinary synchronous and streaming inference does not retain it. The opt-in Batch API temporarily retains enclave-encrypted artifacts for up to 30 days. You can verify the gateway yourself: the attestation endpoint returns a JWT signed by the CPU vendor's root key, bound to your live TLS session.
How do I switch from OpenRouter to TrustedRouter for MiniMax M3?
Change your base_url to the TrustedRouter endpoint and keep your SDK, your code, and the minimax/minimax-m3 model id. The API is OpenAI-compatible. Automatic provider fallback rolls requests over when a provider is down, and measured latency from real routed requests is published on the latency benchmarks page.