ERNIE 5.5 Flash Preview is cheap to call — Qianfan peak ¥1/¥4 and off-peak ¥0.5/¥2 after the limited-time 50% discount (original ¥2/¥8), OpenRouter $0.28/$1.13, and a $0 free tier. This article settles three accounts first: Qianfan's peak/off-peak tiers at the 50%-off rate, OpenRouter's dollar pricing and free-tier limits, and which channel fits which user. Prices checked Nov 17, 2026 (51AllAI, citing Qianfan's model-service billing page); the checkout page price is authoritative.
Baidu AI Cloud's Qianfan API (official pricing, live): the original price is ¥2 input / ¥8 output per 1M tokens; after the limited-time 50% discount, peak hours cost ¥1 in / ¥4 out and off-peak hours ¥0.5 in / ¥2 out; peak hours are 8:00–22:00 and off-peak hours 22:00–8:00 the next day; cache-hit input after the discount is ¥0.1 at peak and ¥0.05 off-peak. Model ID ernie-5.5-flash-preview, listed for both text generation and deep thinking; default rate limits are 60 RPM and 1,000,000 TPM. One easily missed point: reasoning is a fixed high-effort tier with no toggle, so output tokens on long tasks are harder to compress than on models with an adjustable tier — budget the output line for high effort. The wording is per 51AllAI, citing Qianfan's model-service billing page; the Qianfan console is authoritative.
Two readings used to coexist: aiho.net and todayforai only reported "the vendor says limited-time 50% off" without stating whether ¥2/¥8 was the discounted price, while yunsilu.net said "peak ¥1/¥4 after the discount" (i.e., ¥2/¥8 is the original price). 51AllAI, citing Qianfan's model-service billing page, has settled it: ¥2/¥8 is the original price, and the limited-time 50% discount brings peak hours to ¥1/¥4 and off-peak hours to ¥0.5/¥2, with peak hours 8:00–22:00 and off-peak hours 22:00–8:00 the next day. Budget on the discounted rate, but keep one caveat in mind: the discount is time-limited, so check the Qianfan console for its end date — prices may revert to the original rate when it ends.
OpenRouter (listed Nov 10): the paid tier baidu/ernie-5.5-flash-preview costs $0.28 input / $1.13 output per 1M tokens, with cache reads at $0.028; context 1,024,000. The free tier baidu/ernie-5.5-flash-preview:free costs $0, limited to 20 requests per minute and 50 requests per day; buying 10 credits raises it to 1,000/day (verified by apidog on Nov 12). Converted, the paid OpenRouter tier ($0.28/$1.13) is roughly Qianfan's original peak rate (¥2/¥8) — but Qianfan's discounted peak rate (¥1/¥4) is about half of that, so Qianfan is cheaper while the 50% discount lasts; the choice is mostly about billing currency and existing channels, not unit price alone. The free tier's 50 requests a day are enough to try it: validate the output, run a few agent tasks, compare the style once. Not enough to run it as a workhorse — after 50, wait for the next day, buy 10 credits for 1,000/day, or switch to a paid tier.
| Scenario | Recommended channel | Why |
|---|---|---|
| First look at the output | Web chat at ernie.baidu.com or OpenRouter free tier | $0; 50 free requests a day are enough to validate |
| China-based, paying in RMB | Qianfan API | Official channel; peak/off-peak tiers, ¥0.5/¥2 off-peak after the 50% discount |
| Paying in USD, already on OpenRouter | OpenRouter paid tier | $0.28/$1.13; cache reads $0.028 |
| High-frequency agent work | Qianfan or OpenRouter paid tier | 50 free requests/day is not enough; fixed high-effort reasoning pushes the output bill up |
| Long documents (1M context) | Either paid channel | Input tokens are large, the free tier's limits hit first; max output 921,600 tokens — budget on the output price |
One reminder: a 1M context sounds large, but billing is per token — input tokens on a long-document task are themselves a cost, and output tokens (fixed high-effort tier) are a second one. Run one real task on the free tier first and measure its actual token use before picking a paid channel; that beats comparing unit prices.
No. 51AllAI, citing Qianfan's model-service billing page, confirms that ¥2/¥8 is the original price: after the limited-time 50% discount, peak hours cost ¥1 in / ¥4 out and off-peak hours ¥0.5 in / ¥2 out; peak hours are 8:00–22:00 and off-peak hours 22:00–8:00 the next day; cache-hit input after the discount is ¥0.1 at peak and ¥0.05 off-peak. The Qianfan console is authoritative, and the discount is time-limited — check its end date when budgeting.
OpenRouter's free tier allows 20 requests per minute and 50 requests per day; buying 10 credits raises it to 1,000/day (verified by apidog on Nov 12).
Qianfan, while the discount lasts: the paid OpenRouter tier ($0.28/$1.13) converts to about Qianfan's original peak rate (¥2/¥8), but Qianfan's discounted peak (¥1/¥4) and off-peak (¥0.5/¥2) rates are roughly half of OpenRouter's; peak hours are 8:00–22:00 and off-peak hours 22:00–8:00 the next day. Once the discount ends, the two are roughly the same — pick by billing currency and existing channels.
Reference: Baidu official release page — ERNIE 5.5 Flash Preview · OpenRouter — baidu/ernie-5.5-flash-preview:free · aiho.net — 280B deep dive · 51AllAI — ERNIE 5.5 Flash Preview deep dive (citing Qianfan's model-service billing page) (rewritten and organized by ModelBrief; facts per the official release page, Qianfan docs and OpenRouter)