mimo-v2-flash Migration: Xiaomi API Alias Now Routes to MiMo V2.5
Searching for mimo-v2-flash or the Xiaomi MiMo V2 Flash API now means checking a migration alias, not adopting a fresh launch model. Xiaomi started system-replacing mimo-v2-flash with mimo-v2.5 on June 18, 2026, and the V2 model-name cutoff was June 30. If your TheRouter routing policy still pins xiaomi/mimo-v2-flash, treat this page as a migration checklist: confirm the effective model, compare pricing, and retest thinking-mode parameters before you rely on old Flash behavior.
What changed after June 18
Xiaomi published a retirement schedule: requests sent as mimo-v2-flash are accepted during the grace window but are automatically served by mimo-v2.5. The old ID then reaches hard deprecation on June 30, 2026. That means operators should not evaluate this route as a stable V2 Flash endpoint anymore.
- Use explicit V2.5 IDs for new configs. Prefer
xiaomi/mimo-v2.5where you want the replacement behavior to be visible in logs, dashboards, and routing rules. - Rebaseline cost. The legacy Flash launch pricing and the replacement V2.5 pricing are not the same commercial object; check your TheRouter usage ledger after switching.
- Retest thinking mode. V2.5 handles thinking-mode defaults and temperature/top_p behavior differently from V2 Flash in Xiaomi's migration notes, so eval snapshots can drift even when calls still succeed.
For the detailed vendor timeline, see our deprecation brief: MiMo V2 Flash deprecation and V2.5 migration.
How prompt-cache billing worked at launch
At launch, xiaomi/mimo-v2-flash was listed on TheRouter through the xiaomi route with first-class prompt-cache billing. When the upstream response surfaced usage.prompt_tokens_details.cached_tokens, the meter split prompt-token cost into two buckets:
- Non-cached tokens at the standard
inputrate. - Cached tokens at the
cache_readrate. - Output tokens at the standard output rate.
The migration does not require client SDK changes, but it does change what you should measure. Watch cache-hit rate, effective input cost, output cost, and any reasoning-token usage after moving traffic to xiaomi/mimo-v2.5.
Pricing & capability checklist
| Item | What to verify |
|---|---|
| Legacy alias | xiaomi/mimo-v2-flash may be silently remapped upstream during the grace window; avoid using it as a durable production target. |
| Replacement target | Pin xiaomi/mimo-v2.5 explicitly for the post-migration route. |
| Context | Confirm current catalog values before a large batch run. |
| Capabilities | Streaming · tools · JSON mode remain the integration baseline. |
| Operational risk | Silent alias drift: successful calls can hide behavior and billing changes. |
Recommended TheRouter config
curl https://api.therouter.ai/v1/chat/completions -H "Authorization: Bearer $THEROUTER_API_KEY" -H "Content-Type: application/json" -d '{
"model": "xiaomi/mimo-v2.5",
"messages": [
{"role": "system", "content": "<long stable system prompt>..."},
{"role": "user", "content": "Question 1"}
],
"provider": {"only": ["xiaomi"]}
}'If you still have historical traffic under xiaomi/mimo-v2-flash, run a small canary on xiaomi/mimo-v2.5 and compare quality, cost, latency, and parameter behavior before shifting all routes.