Loading
90 days of the enterprise, pro and team plans free, no card needed. See the plans
Model comparison
Qwen's Qwen3.8 2.4T A95B against Xiaomi's MiMo-V2.6-Pro: vendor list prices, limits, abilities and the Artificial Analysis indexes side by side - and what OpenSmartRoute does when both are available.
MiMo-V2.6-Pro lists at $0.544 per million tokens blended against $3.00 for Qwen3.8 2.4T A95B - 5.5x less.
MiMo-V2.6-Pro scores 46 on the intelligence index against 40 - 6 points apart.
MiMo-V2.6-Pro reads 1.1M tokens at once, Qwen3.8 2.4T A95B 1.0M.
only MiMo-V2.6-Pro offers image input.
MiMo-V2.6-Pro was released about 1 month later.
Vendor list prices per million tokens and published limits; indexes are Artificial Analysis composites (0-100).
| Specification | Qwen3.8 2.4T A95B | MiMo-V2.6-Pro |
|---|---|---|
| Input price / 1M tokens | $2.00 | $0.43 |
| Output price / 1M tokens | $6.00 | $0.87 |
| Blended (3 in : 1 out) | $3.00 | $0.54 |
| Cached input / 1M | $0.25 | $0.004 |
| Context window | 1.0M | 1.1M |
| Max output | 909K | 131K |
| Input | text | text, image, video, audio |
| Output | text | text |
| Tool calls | ||
| Structured output | ||
| Reasoning | ||
| Open weights | ||
| Intelligence index | 40 | 46 |
| Coding index | 72 | - |
| Agentic index | 50 | - |
| Released | Aug 12, 2026 | Sep 21, 2026 |
| Available on this deployment |
List price times tokens for three typical workloads; your mix of input and output decides which model's pricing suits you.
| Workload | Qwen3.8 2.4T A95B | MiMo-V2.6-Pro | Difference |
|---|---|---|---|
| Support chat20M tokens in, 5M out a month - a help desk answering a few thousand conversations. | $70.00 | $13.05 | $56.95 less on MiMo-V2.6-Pro |
| Document summaries100M in, 5M out - long inputs, short answers. | $230 | $47.85 | $182 less on MiMo-V2.6-Pro |
| Coding assistant60M in, 20M out - a team's editor assistant, output-heavy. | $240 | $43.50 | $197 less on MiMo-V2.6-Pro |
MiMo-V2.6-Pro is both cheaper and stronger on the intelligence index, so with both in the catalogue the router sends it almost everything; Qwen3.8 2.4T A95B earns a request only on something it alone offers - a longer context, a modality, a data boundary or a region your rules require.
# targets.yaml - both models as routing targets (metadata.model links each to the catalogue entry)
targets:
- { id: mimo-v2.6-pro, kind: llm, metadata: { model: xiaomi/mimo-v2.6-pro } }
- { id: qwen3.8-2.4t-a95b, kind: llm, metadata: { model: qwen/qwen3.8-2.4t-a95b } }
# rules.yaml - the cheaper model by default, the stronger one when the request is hard
rules:
- name: simple-requests-stay-cheap
when: { max_complexity: 0.45 }
prefer: [mimo-v2.6-pro]
weight: 0.8
- name: hard-requests-escalate
when: { min_complexity: 0.45 }
prefer: [qwen3.8-2.4t-a95b]
weight: 0.8Every Qwen and Xiaomi model, with prices, is on the vendor pages; the rankings show what the router actually picks.