Yomi API

DeepSeek has recently adjusted the pricing and billing rules for API models, introducing peak/off-peak billing.

To align with upstream official prices, Yomi API will adjust DeepSeek model prices starting August 17, 2026, covering input, cache hit, and output token billing for DeepSeek V4 series models.

This adjustment only syncs upstream official prices and does not affect other models or services. Specific prices are subject to the Yomi API model plaza and pricing page.

Thank you for your understanding and support.

The discount multiplier for Zhipu domestic models has been reduced from 0.6 to 0.35, effective immediately. Existing keys and call methods require no changes. Thank you for your support.

Taking GPT-5.6-sol as an example, the correct model name is: gpt-5.6-sol Incorrect examples: gpt5.6, gpt5.6sol, entering API Key in the model name field, etc. Model names must exactly match the Model ID provided by the platform. Do not abbreviate, modify, or guess. The Yomi 'Model Services' page provides accurate model names. Use the names as displayed, copy directly. Model name ≠ API Key, do not confuse them.

Due to upstream model service adjustments, the GPT-5.6 Luna model will be temporarily suspended starting August 6, 2026 at 10:00 Beijing time. At that time, the model will be removed from the Yomi API model list, and users will no longer be able to call gpt-5.6-luna. If your application or project is using this model, please complete migration in advance to avoid disruption. Yomi API will continue to monitor model service updates and provide stable, reliable alternatives in a timely manner. Thank you for your understanding and support.

Single recharge of 100 yuan gives 20 yuan bonus, 500 yuan gives 120 yuan, 1000 yuan gives 300 yuan. Bonus balance is credited instantly with no expiration.

The help documentation has been completely restructured, adding a quick start guide, media knowledge base, and FAQ quick reference.

Single recharge of 100 yuan or more gets an extra 10% bonus. The event deadline is as shown on the recharge page.

This upgrade covers four core nodes: Tokyo, Singapore, Frankfurt, and US East. The gateway has been switched to a multi-instance load balancing architecture.

After the upgrade, P99 latency decreased from 840ms to 490ms, and long-connection disconnection rate dropped by 61%. Existing keys require no adjustment; all traffic has been automatically switched to the new gateway.

The announcement center now supports category filtering and multiple display formats. The help documentation now includes full-text search; enter keywords to locate specific content paragraphs.

Qwen3 Max and DeepSeek R2 have been integrated into the platform's unified catalog. Qwen3 Max focuses on general conversation and long text, while DeepSeek R2 enhances reasoning and mathematical capabilities. Both support all client integrations, and existing keys require no adjustment.

Mid-year rewards event: Single recharge of 100 gets 10% bonus, 500 gets 20%, 1000 gets 30%. Bonus balance is credited instantly, valid for all models, with no expiration.

Three new access points have been added: Frankfurt, Silicon Valley, and São Paulo, bringing the total global nodes to 12. Average access latency for European and South American users has decreased by approximately 35%, with automatic optimal routing and no configuration changes required.

From 02:17 to 02:29 today, the US East node experienced network jitter. The gateway's auto-retry mechanism was activated, no failed requests occurred, and service has fully recovered.

The console usage page has been fully upgraded: added usage breakdown by model, API key, and time dimension, with CSV export support. Cost attribution is now clear at a glance.

Claude Sonnet 4.5 has been fully integrated, with stable performance in coding, tool calling, and long-context scenarios. Platform prices are lower than official catalog prices and support all major clients.

Invite new users to register and recharge, and you get 8% cashback of their recharge amount. Friends also enjoy a 5% discount on their first recharge. Invitation links and cashback details are on the 'Referral Rewards' page in the console.

Gemini 2.5 Pro has been integrated into the platform, supporting ultra-long context and multimodal input such as images and audio. It uses the same catalog and billing as existing models.

The platform's daily average API calls have exceeded 120 million, with overall availability maintained above 99.95%. Thank you for the trust of every developer. We will continue to improve stability.

API keys now support setting monthly quota limits and automatic expiration per key, allowing precise cost control even with shared team accounts.