MiMo-V2.6-Pro on QuickSilver Pro
MiMo-V2.6-Pro is Xiaomi's flagship open-weight model — a 1.02T-parameter mixture-of-experts with 42B active — and the top open-weights model on the Artificial Analysis Intelligence Index v4.3, scoring 46 (MiMo-V2.5-Pro scored 26). It accepts text and images across a 1M-token context with streaming, tools, and structured output. QuickSilver Pro charges $0.348 input / $0.696 output per million tokens — 20% below Xiaomi's $0.435 / $0.87 list price on both legs.
At a glance
The strongest open-weights model for coding and agents, with image input and a 1M-token context.
Pricing comparison ($/1M tokens)
| Provider | Input | Output | vs QSP |
|---|---|---|---|
| QuickSilver Pro | $0.348 | $0.696 | lowest-cost |
| Xiaomi list price (Xiaomi MiMo API) | $0.435 | $0.87 | 20% lower |
When to use
Reach for MiMo-V2.6-Pro when you want frontier-class coding and agentic behavior from an open-weights model at a fraction of closed-model prices: multi-step tool-use agents, repo-scale code generation and review, long-document work over a 1M-token working set, and screenshot or diagram understanding. On Artificial Analysis it sits on the intelligence-versus-cost-per-task Pareto frontier, so it is a strong default when you are optimizing quality per dollar.
When to use something else
For high-volume, latency-sensitive traffic where the flagship is more than you need, MiMo-V2.6-Flash ($0.112/$0.224) is about a third of the price. For the very hardest reasoning and long-horizon agent work, closed frontier models such as Claude Opus 5 or GPT-6 Astra still score higher — at many times the per-token price. On QuickSilver Pro it runs in non-thinking mode, which keeps output short and cheap.
Quickstart (curl)
curl https://api.quicksilverpro.io/v1/chat/completions \
-H "Authorization: Bearer $QSP_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "mimo-v2.6-pro",
"messages": [{"role": "user", "content": "Hello!"}]
}'OpenAI-compatible. One-line migration via base_url.
FAQ
$0.348 per million input tokens, $0.696 per million output tokens, and $0.00288 per million cached-input tokens. That is 20% below Xiaomi's list price of $0.435 input / $0.87 output ($0.0036 cached) — lower on both input and output, so the saving holds for output-heavy coding agents too.
It is a reasoning-capable model, but QuickSilver Pro serves it in non-thinking mode: reasoning is off, so calls return a direct answer without billing a hidden reasoning trace. A `reasoning` field in the request does not turn it back on today. If a task needs step-by-step reasoning, use a model that reasons by default, such as Kimi K3 or DeepSeek V4 Pro.
On the Artificial Analysis Intelligence Index v4.3 it scores 46, the highest of any open-weights model at release (2026-09-21) and well ahead of MiMo-V2.5-Pro's 26. It also sits on AA's intelligence-versus-cost-per-task Pareto frontier. As always, run it against your own evals before switching production traffic.
Set base_url=https://api.quicksilverpro.io/v1, use your QSP API key, and set model="mimo-v2.6-pro". Both `/v1/chat/completions` and `/v1/responses` are supported, along with streaming, tool calling, json_schema strict mode, image input, and usage.cost accounting.