MiMo V2 Flash Deprecation: June 18 Migration to V2.5 Changes Your API Routing

The decision that changes your MiMo API costs this week isn't whether to migrate — Xiaomi has already made that choice for you. The question is whether your routing config, billing dashboards, and parameter assumptions are ready for June 18.

What happened

Xiaomi's MiMo platform published its V2 model retirement schedule. Two deadlines matter now:

June 18, 2026, 00:00 Beijing Time (system replacement): Starting at this moment, any request sent with model: mimo-v2-flash or model: mimo-v2-tts will be automatically rerouted to mimo-v2.5 and mimo-v2.5-tts respectively. The old model ID is accepted but silently remapped. Billing switches to V2.5 rates immediately.

June 30, 2026, 00:00 Beijing Time (hard deprecation): After this point, calls using V2 model names return an error. There is no more grace period.

The two earlier models (mimo-v2-pro and mimo-v2-omni) already passed their June 1 system-replacement date — they have been routing to mimo-v2.5-pro and mimo-v2.5 since June 1. For background on the V2.5 model family's capabilities and API surface, see our coverage of the MiMo V2.5 provider launch and the V2.5-Pro UltraSpeed inference update.

The full deprecation table:

Why it matters for AI engineering teams

Pricing surprise. If you hardcoded mimo-v2-flash in your routing rules because it had a specific price tier, that tier changes on June 18 without any API error telling you so. Requests succeed but are billed at V2.5 rates. You won't see this in model-name logs — everything still looks like mimo-v2-flash until you check the billing detail page.

Silent parameter rewriting for mimo-v2-flash. This is the subtler risk. When mimo-v2-flash auto-routes to mimo-v2.5, the parameter handling diverges in thinking mode:

• V2.5 does not support custom temperature or top_p in thinking mode. If your request includes these, they will be ignored and clamped to temperature: 1.0, top_p: 0.95.
• If thinking, temperature, or max_completion_tokens are not specified, the system substitutes V2.5 default values — which differ from V2-Flash defaults.

This means eval pipelines or cost-controlled inference workloads that relied on specific temperature settings in thinking mode will silently shift behavior starting June 18.

TTS timbre remapping. For mimo-v2-tts users: the default voice mimo_default is remapped — to 冰糖 in Chinese-locale clusters and mia in all other clusters. If your voice output is locale-sensitive, test your TTS integration before June 18.

The router/operator angle

This deprecation pattern — auto-route first, hard kill later — is how every mature AI provider manages model lifecycle. The operational risk isn't the migration itself; it's the asymmetry between what your routing config says and what is actually running.

Four things to do before June 18:

1. Audit model IDs in your router config. Any route that still says mimo-v2-flash or mimo-v2-tts will silently change behavior and pricing at midnight Beijing time on June 18. Update to explicit V2.5 model names (mimo-v2.5, mimo-v2.5-tts) so the behavior is intentional, not accidental.

2. Check your billing baseline now. Visit the MiMo console bill details to see current spend by model. Use this as your pre-migration baseline so you can compare post-June-18 numbers immediately.

3. Review thinking-mode parameter usage. If any mimo-v2-flash call includes explicit temperature or top_p alongside thinking: enabled, those parameters will stop taking effect. Adjust your request payloads before June 18 or accept V2.5 defaults.

4. Test TTS voice output if you use mimo_default. Switch to an explicit voice name in your TTS config so the remapping doesn't affect production output.

Provider dependency review. If you run MiMo V2 on a third-party aggregator (SiliconFlow, OpenRouter, or a hosted gateway), check whether that aggregator's model aliases have also been updated. Some aggregators apply their own model-name mapping layer — your upstream alias may still point to V2 on their side even after you update your config, or vice versa.

What TheRouter users should watch or try

If you have configured MiMo as a provider in TheRouter with explicit V2 model IDs (xiaomi/mimo-v2-flash, xiaomi/mimo-v2-tts), update your routing rules to use V2.5 model names before June 18. This makes the migration explicit, auditable in your routing logs, and decoupled from Xiaomi's silent remapping behavior.

For billing monitoring: after June 18, cross-reference your TheRouter usage dashboard against the MiMo console bill detail page if you're running high MiMo flash volume. V2.5 pricing should be reflected in both.

The broader pattern to watch: providers that run a "system replacement" phase before hard deprecation are giving you a window to validate V2.5 output quality on your actual workloads before the fallback disappears entirely. Use that window.