Wink Pings

Hy4 flash is 18x cheaper than Claude Haiku—first, let's look at how the blind test was conducted

Tencent Hunyuan releases Hy4 flash; the company says it matches Claude Haiku 5.5 high on everyday tasks and cuts task cost by 18x. API pricing is $0.07 per million input tokens, $0.003 for cache, and $0.281 for output. Beyond the low price, the blind-test details and availability channels became the focus in the replies.

Tencent Hunyuan released Hy4 flash on November 16, positioning it as the cheapest model for everyday productivity scenarios. The official comparison target is Claude Haiku 5.5 (high), and the claim is that it ties in human blind tests on everyday work tasks while reducing task cost by 18x. First, pricing: per million API tokens, cache hits cost $0.003, input costs $0.070, and output costs $0.281. At this pricing, batch tasks and repeated context will clearly affect the bill. Official research page: https://hy.tencent.ai/research/hy4-flash

image

ItemPer million tokens
Cache hit$0.003
Input$0.070
Output$0.281

image

Low price is obvious. The official tweet did not go into the blind test’s test set, sample size, or scoring method. Some users wrote in the replies that the blind test needs to support the “cheapest” claim. That criticism isn’t harsh; the more aggressive the pricing, the more the testing methodology needs to be laid out. Another practical issue is availability channels. Some asked whether it could be on OpenRouter, some asked about open weights, and others mentioned that Hy4 Flash Free already appears in OpenCode’s model list. For developers, whether they can call it directly in the tools they already use is more concrete than the 18x on the poster.

image

image

Reactions from users fell into a few categories: some highlighted the 18x cost reduction, some said that rounded off it’s basically free, and some asked whether the translation model HY-MT was about to get an update. The official tweet did not respond to these. For ordinary users, API pricing is per million tokens, with cache hits, input, and output billed separately; the actual cost depends on the task structure. Hy4 flash pushes the price very low, but whether to switch depends first on whether your tasks fall within what the company calls everyday productivity, and then on the blind-test details and integration channels.

发布时间: 2026-11-16 17:26