Xiaomi Permanently Cuts MiMo-V2.5 Prices by Up to 99%, Matching DeepSeek V4 Pro Rates

What changed in MiMo-V2.5 pricing

The price reduction covers both MiMo-V2.5 and MiMo-V2.5 Pro and took effect globally at 00:00 Beijing Time on May 27 . The old V2 model line remains at its existing prices and is marked for deprecation .

Xiaomi's official pricing page confirms three billing tiers per model—cache hit, cache miss, and output—measured per million tokens . The overseas rates for the MiMo-V2.5 Pro now stand at:

• Input (cache hit): $0.0036 per million tokens
• Input (cache miss): $0.435 per million tokens
• Output: $0.87 per million tokens

The base MiMo-V2.5 overseas pricing is even lower:

• Input (cache hit): $0.0028 per million tokens
• Input (cache miss): $0.14 per million tokens
• Output: $0.28 per million tokens

Those cache-hit figures are where the 99% reduction is most visible. Since many production workloads reuse the same system prompts or document prefixes, developers who structure their applications to maximize cache hits can see costs drop by an order of magnitude or more . A Hacker News commenter estimated that a workload that previously cost roughly $400 at the old MiMo pricing would cost around $40 after the cut .

How MiMo-V2.5 Pro stacks up against DeepSeek V4 Pro

Multiple sources note that MiMo-V2.5 Pro's new cache-miss rate of $0.435 per million tokens matches DeepSeek V4 Pro's current promotional pricing . DeepSeek V4 Pro is currently offered at a 75% discount through May 31, 2026, which sets the promo rates at $0.435 per million input tokens and $0.87 per million output tokens . After the promotion, V4 Pro's standard list price rises to $1.74 input and $3.48 output .

DeepSeek also publishes separate Flash and Pro tiers with identical 1M-token context windows and up to 384K output tokens per request . V4 Flash costs $0.14 per million input tokens (cache miss) and $0.28 per million output tokens, making it competitive with MiMo-V2.5's base tier .

The practical alignment between MiMo-V2.5 Pro and DeepSeek V4 Pro on current pricing is as close as it gets for frontier-model access in China as of late May 2026. The key difference is that Xiaomi's pricing is permanent with no stated expiration date, while DeepSeek's promo rate expires after May 31 .

The engineering rationale behind the cuts

For DeepSeek, the engineering story is documented directly. DeepSeek V4 was optimized to run on domestic Huawei Ascend 950 chips, which allowed the company to price API access dramatically below competing frontier models from OpenAI and Anthropic . Counterpoint Research Vice President Neil Shah described the pricing as "a serious flex" on inference cost, and Principal Analyst Wei Sun noted V4's "excellent agent capability at significantly lower cost" .

Xiaomi's official announcement frames the price cut in terms of passing "technological dividends" to developers and reducing calling costs through better billing system optimization, particularly by eliminating context-length-based price tiers . The available sources from Xiaomi itself do not provide detailed engineering disclosures about chip optimization or inference efficiency gains in the same depth as DeepSeek's documentation, but they make clear the company is competing directly on API cost .

The other side: Zhipu AI and rising prices

While Xiaomi and DeepSeek are cutting prices, several major Chinese AI firms have moved in the opposite direction during early 2026.

Zhipu AI raised API prices by 83% in the first quarter of 2026, following earlier increases of 30% in February and a further 10% in April . Despite the price hikes, Zhipu's call volumes rose rather than fell—jumping approximately 400% according to multiple sources . CEO Zhang Peng emphasized that high cost-performance, in Zhipu's view, does not mean cutthroat price competition .

Moonshot AI raised Kimi input-token prices by 58% when it launched the K2.6 model in late April, moving the API input price from $0.60 to $0.95 per million tokens . The company's annual recurring revenue surpassed $200 million in April alone, driven by rapid growth in paid subscriptions and API usage . Tencent Cloud raised prices on its Hunyuan model series by more than 400% and ended free public beta access for several third-party models .

The split in pricing strategy reflects a broader divergence in China's AI market. Companies focused on high-end reasoning and commercial products are finding that demand is sufficiently inelastic to support price increases, while compute-optimized firms like DeepSeek and Xiaomi are competing for volume and developer share through low-cost access .