Faster turns
Each answer streams back sooner
PrimaLabs is #1 of 5 MiMo-V2.6-Pro providers for output speed on Artificial Analysis: 199.8 tokens per second, against 48.2 for the next provider.
MiMo-V2.6-Pro API, #1 on Artificial Analysis for speed
#1 of 5 providers for output speed and latency
PrimaLabs runs Xiaomi's MiMo-V2.6-Pro language model as a serverless inference API: OpenAI-compatible, 1M-token context, billed per token.
$0.435 input · $0.87 output per 1M tokens · pay as you go, no upfront commitment
Measured by Artificial Analysis, not by us
Independently benchmarked: Artificial Analysis ranks PrimaLabs #1 of 5 MiMo-V2.6-Pro API providers on both output speed and time to first answer token. Recent readings reach up to 216 tokens per second; against the next provider, PrimaLabs is at least 3x faster on both output speed and first answer.
199.8
output tokens per second
#1 of 5 · Artificial Analysis10.72 s
until the first answer token arrives
#1 of 5 · Artificial AnalysisSource: Artificial Analysis, MiMo-V2.6-Pro provider benchmark
Built for coding agents and long-context work
A coding agent, like any agentic workflow, calls the model in a loop: each turn resends the conversation and waits for a long answer. Two things decide how that loop feels and what it costs.
Cheaper long loops
History the cache holds is billed as cached input
When a request starts with a prefix the cache already holds, those tokens cost $0.0036 per 1M instead of $0.435, 99% less. An agent resends the same history every turn, so that history is what the cache can serve.
- OpenAI-compatible Change the base URL, API key and model; streaming works the same.
- 1M-token context 1,048,576 tokens: a large codebase or a long document in one request.
- Zero data retention PrimaLabs does not retain your prompts or outputs.
MiMo-V2.6-Pro API pricing
MiMo-V2.6-Pro on PrimaLabs is billed per token, at the same per-token rates as Xiaomi's standard tier: pay as you go, with no upfront commitment.
| Token type | Price per million tokens | When it applies |
|---|---|---|
| Input | $0.435 | New prompt tokens |
| Output | $0.87 | Tokens the model generates |
| Cached input | $0.0036 | Prompt tokens served from the prefix cache |
- Input: every prompt token that is not served from the cache.
- Cached input: prompt tokens at the start of a request that match a prefix the cache holds, such as an agent's history or a long system prompt. They bill 99% below the input rate.
- Output: every token the model generates, streamed or not.
Example agent turn: 100,000 tokens of history served from the cache, 2,000 new input tokens and a 1,500-token answer cost $0.0025. Billed as fresh input, the same history would make that turn $0.046.
Example computed from the rates above. Your dashboard always shows your current rates and usage.
From API key to first response in under a minute
Existing OpenAI clients change three things: the base URL, the API key and the model, set to primalabs-ai/MiMo-V2.6-Pro.
- Create an API key in the dashboard.
- Point your OpenAI SDK at
https://api.primalabs.ai/v1. - Send a request and get the answer back, streamed or in one piece.
Python
from openai import OpenAI
client = OpenAI(
base_url="https://api.primalabs.ai/v1",
api_key="YOUR_API_KEY",
)
r = client.chat.completions.create(
model="primalabs-ai/MiMo-V2.6-Pro",
messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)Node
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.primalabs.ai/v1",
apiKey: "YOUR_API_KEY",
});
const r = await client.chat.completions.create({
model: "primalabs-ai/MiMo-V2.6-Pro",
messages: [{ role: "user", content: "Hello" }],
});
console.log(r.choices[0].message.content);cURL
curl -sS https://api.primalabs.ai/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "primalabs-ai/MiMo-V2.6-Pro",
"messages": [{"role": "user", "content": "Hello"}]}'
Response, streamed (illustration)
Hello. How can I help you today?
MiMo-V2.6-Pro API endpoint and access
Access is self-serve: create an API key in the dashboard and call the endpoint from the SDK you already use.
| Base URL | https://api.primalabs.ai/v1 |
|---|---|
| Model | primalabs-ai/MiMo-V2.6-Pro |
| Authentication | Authorization: Bearer YOUR_API_KEY |
| Endpoints | POST /chat/completions (primary) · POST /responses (OpenAI Responses format) · POST /messages (Anthropic Messages format) · POST /completions (legacy) |
| Streaming | stream: true returns tokens as server-sent events |
| Context window | 1,048,576 tokens |
Streaming request, as in the docs quickstart:
curl -sS -N https://api.primalabs.ai/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "primalabs-ai/MiMo-V2.6-Pro", "stream": true,
"messages": [{"role": "user", "content": "Hello"}]}'Choosing a MiMo-V2.6-Pro inference provider
Five MiMo-V2.6-Pro API providers are on Artificial Analysis, and they show the same blended price there. Compare them on speed, latency, context, data handling and room to grow; here is PrimaLabs on each.
- Output speed
- #1 of 5 at 199.8 tokens per second: the fastest-ranked MiMo-V2.6-Pro endpoint on Artificial Analysis.
- Latency
- #1 of 5 for time to first answer token, at 10.72 s: lower than the other four providers on Artificial Analysis.
- Price
- Standard per-token rates, with cached input at $0.0036 per 1M tokens.
- Context
- A 1,048,576-token context window.
- Data handling
- Zero data retention on serverless.
- Room to grow
- Dedicated GPUs available when traffic is steady and high.
MiMo-V2.5 deprecation: switch to MiMo-V2.6-Pro
Moving off MiMo-V2.5? Xiaomi retires (deprecates) mimo-v2.5-pro and mimo-v2.5 on its API on 21 Oct 2026. As a mimo-v2.5-pro alternative, MiMo-V2.6-Pro on PrimaLabs is a three-line switch: the base URL, the API key and the model.
client = OpenAI(
base_url=OLD_BASE_URL,
base_url="https://api.primalabs.ai/v1",
api_key=OLD_API_KEY,
api_key=PRIMALABS_API_KEY,
)
client.chat.completions.create(
model="mimo-v2.5-pro",
model="primalabs-ai/MiMo-V2.6-Pro",
messages=history,
)Dedicated capacity
Dedicated GPUs for steady, high-volume MiMo workloads
For enterprise deployment, PrimaLabs runs MiMo-V2.6-Pro on GPUs dedicated to your workload, available on request. We benchmark it head to head and put throughput, latency and cost in writing before any commitment.
- Share your workload
- Head-to-head benchmark
- Results in writing
MiMo-V2.6-Pro API questions
How much does the MiMo-V2.6-Pro API cost?
$0.435 per 1M input tokens, $0.87 per 1M output tokens and $0.0036 per 1M cached input tokens, billed per token: pay as you go, with no upfront commitment. Input served from the prefix cache is billed at the cached rate, 99% below the input rate. Your dashboard always shows the current rates.
How do I call MiMo-V2.6-Pro on PrimaLabs?
Create an API key in the dashboard, point any OpenAI SDK at https://api.primalabs.ai/v1 and set the model to primalabs-ai/MiMo-V2.6-Pro. Existing OpenAI clients change three things: the base URL, the API key and the model.
How fast is MiMo-V2.6-Pro on PrimaLabs?
Artificial Analysis ranks PrimaLabs #1 of 5 MiMo-V2.6-Pro providers for output speed, at 199.8 tokens per second. It also ranks PrimaLabs #1 for time to first answer token, at 10.72 s.
Is this the MiMo-V2.6-Pro UltraSpeed edition?
No. UltraSpeed is a separate, higher-priced edition from Xiaomi. As an alternative at standard rates, PrimaLabs serves the standard MiMo-V2.6-Pro, the edition Artificial Analysis benchmarks: there PrimaLabs is #1 of 5 providers for output speed.
Does PrimaLabs keep my prompts or outputs?
No. MiMo-V2.6-Pro on PrimaLabs runs with zero data retention: PrimaLabs does not retain your prompts or outputs after serving your request.
What is MiMo-V2.6-Pro?
MiMo-V2.6-Pro is a large language model (LLM) from Xiaomi. PrimaLabs serves it as primalabs-ai/MiMo-V2.6-Pro on a serverless, OpenAI-compatible API with a 1,048,576-token context window.
Xiaomi is retiring MiMo-V2.5. What should I do?
Xiaomi retires mimo-v2.5-pro and mimo-v2.5 on its API on 21 Oct 2026. To move to MiMo-V2.6-Pro on PrimaLabs, change the base URL to https://api.primalabs.ai/v1, use your PrimaLabs API key and set the model to primalabs-ai/MiMo-V2.6-Pro.
Can I run MiMo-V2.6-Pro on dedicated capacity?
Yes. For steady, high-volume workloads, PrimaLabs runs MiMo-V2.6-Pro on dedicated GPUs. We benchmark your workload head to head and put throughput, latency and cost in writing before any commitment.
Point your OpenAI SDK at MiMo-V2.6-Pro today
Billed per token. Pay as you go, no upfront commitment.