MiMo-V2.6 API: Is It Out Yet, and What Can You Use Today?

Is MiMo-V2.6 available as an API?

No. As of September 17, 2026, none of the pieces a developer needs exist:

Item	MiMo-V2.6 status
Public API model ID	Not published — the two run names are mimo-v2.6-pro and mimo-v2.6-flash
Official price	Not published
Model card, context window, parameters	Not published
Release date	Not announced; Xiaomi's guidance is "coming soon"
Training status	In progress — both runs are still training
What exists today	A public live training panel at mimo.xiaomi.com/rl

Anyone publishing a MiMo-V2.6 model ID, price or benchmark table right now is filling in gaps that Xiaomi has not filled. Treat those numbers as invented until Xiaomi publishes them.

What is actually confirmed about MiMo-V2.6

Four things are confirmed, and they come from Xiaomi or from the training panel itself:

• Two variants are being trained: mimo-v2.6-pro and mimo-v2.6-flash, run in parallel since September 15, 2026.
• The training is public. Xiaomi's panel streams steps, pass rates and cost from the trainer's logs — a deliberate move after the MiMo-V2.5 open-source release, and the reason this model is being discussed before launch. The panel opened to the public on September 16, 2026 (UTC).
• The work is reinforcement learning, not a new pre-training run. Xiaomi's team frames it as scaling RL along three axes: training compute, environments and harnesses, and grader compute. Code, general, visual, cyber and chat agent tasks are mixed inside the same run, with code tasks making up roughly two thirds of each step.
• It is not a clean run. The panel itself records restarts, including a VRAM failure on one node for the Pro run and an infrastructure error in the Flash run that was not caught for about three hours. That is worth knowing before quoting any progress curve as a finished result.

The public DeepSWE v1.1 numbers Xiaomi posts alongside the run (mini-swe-agent, avg@3) have moved upward across steps for both variants, and Pro has stayed ahead of Flash. These are Xiaomi's own offline evaluations, published as the results come out — not an independent benchmark, and not a launch spec.

If you want the running numbers, read the panel directly: mimo.xiaomi.com/rl. We deliberately do not copy the cost or token telemetry into this page, because it changes hourly and it is not what you can build on.

MiMo-V2.6 versus MiMo-V2.5: what actually changes for developers

	MiMo-V2.5 (available now)	MiMo-V2.6 (training)
Release status	Released; series open-sourced from April 2026	Not released
Public API	Yes — mimo-v2.5, mimo-v2.5-pro	No public API
Published price	Yes	No
Model card and limits	Published by Xiaomi	Not published
Benchmark standing	Public evaluation results are available	Xiaomi's own in-training evaluations only
What you can do today	Ship on it	Watch the panel, keep your model ID configurable

The practical takeaway: V2.6 does not exist for production use yet. If your roadmap needs a Xiaomi MiMo model this quarter, plan on V2.5 and keep the upgrade to V2.6 a configuration change rather than a rewrite.

What you can use today: MiMo-V2.5 on APIMaster

APIMaster routes mimo-v2.5 and mimo-v2.5-pro through the official Xiaomi channel, with the same per-token rates Xiaomi's model is listed at elsewhere — see the live marketplace card for the current route price, because channel supply changes.

MiMo-V2.5 is a native omnimodal model with a 1M-token context window and a price point that makes it a cheap workhorse for agent loops: at the rates shown on our card, a million input tokens costs cents, not dollars. MiMo-V2.5-Pro sits above it for harder agentic and coding work.

Use your existing APIMaster key and change only the model name:

curl "https://apimaster.ai/v1/chat/completions" \
  -H "Authorization: Bearer $APIMASTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mimo-v2.5",
    "messages": [
      {"role": "user", "content": "Refactor this function and explain the change."}
    ]
  }'

Connecting a Xiaomi coding agent to a third-party key is a separate setup — our MiMo Code guide walks through the base URL and API key screens.

How to prepare for the MiMo-V2.6 launch

1. Keep your model ID configurable. Do not hardcode a guessed mimo-v2.6 string into a client; use a config value you can change in one place.
2. Benchmark the workload, not the model. Pick a representative task from your own product and record correctness, latency and cost per successful task on V2.5 now. That baseline is what makes a V2.6 comparison meaningful later.
3. Watch the official channels. The live panel carries the training status; model IDs and pricing arrive on Xiaomi's platform and, when routes open, on the APIMaster marketplace.
4. Do not plan a migration date. Xiaomi has announced no date, and a training run that is still restarting is not a release candidate.

FAQ

Is MiMo-V2.6 released?

No. MiMo-V2.6 Pro and MiMo-V2.6 Flash are both still in reinforcement-learning training as of September 17, 2026, and Xiaomi has published no release date.

What is the MiMo-V2.6 API model ID?

Xiaomi has not published one. The names visible on the training panel — mimo-v2.6-pro and mimo-v2.6-flash — are training run names, not confirmed public API identifiers.

How much will MiMo-V2.6 cost?

Unknown. Xiaomi has published no price for either variant. Anything quoted today is speculation, including any figure derived from the training panel's cost readout, which measures Xiaomi's own compute spend rather than a future API price.

When will MiMo-V2.6 be available on APIMaster?

When Xiaomi releases it as a public model. We will list the model and its route price on the marketplace; the same APIMaster API key will then cover it, with no new integration.

What is the difference between MiMo-V2.6 Pro and MiMo-V2.6 Flash?

Xiaomi is training both in parallel. Pro is the flagship side of the pair and currently leads Flash on Xiaomi's own public evaluations; Flash is trained for lower cost per token. Neither has a published specification, context window or price yet.

What Xiaomi MiMo model can I use right now?

mimo-v2.5 and mimo-v2.5-pro are available on APIMaster today through the official Xiaomi channel. Check the marketplace for current routes and prices.

Sources and verification date

• Xiaomi's live MiMo-V2.6 training panel: mimo.xiaomi.com/rl — run names, training start, and the status and restart notices cited above.
• Luo Fuli (罗福莉) on X, announcing the public reinforcement-learning stream: x.com/_LuoFuli.
• 量子位 (QbitAI), September 17, 2026: 罗福莉沉寂半年官宣小米强化学习 — the three scaling axes, per-step rollout configuration, and the six-month RL focus.
• BlockBeats, September 2026: 罗福莉憋了半年，小米把 MiMo-V2.6 强化学习直接直播了.
• IT之家 background on the MiMo-V2.5 series: open-source release and Orbit token plan, public beta, MiMo Claw with MiMo-V2.5-Pro.
• APIMaster live pricing catalog and marketplace cards, checked September 17, 2026. Route availability and prices can change after publication.

Start with the MiMo model that exists

MiMo-V2.6 is worth watching, but it is not a model you can call. MiMo-V2.5 and MiMo-V2.5-Pro are.

Create an APIMaster account, generate a key in the console, point your client at https://apimaster.ai/v1 and set the model to mimo-v2.5 — one OpenAI-compatible key, no separate Xiaomi account needed. Every channel is fingerprint-checked before it is listed.