On this page
Tencent Cloud published a pricing notice for the LLM service platform TokenHub and CloudBase: starting 2026-11-06 00:00:00 Asia/Shanghai (Beijing time), Step 5 Preview (model id step-5-preview) will change only its cache-hit input rates. Guangzhou moves 0.35→0.70 RMB per million tokens; Singapore moves 0.36→0.72. The same notice states that cache-miss input and output prices stay unchanged. As of 2026-11-04 Asia/Shanghai—about two calendar days before the effective clock—the adjustment is still a scheduled preview, not a live rate change yet. DigIntoAI has not reconciled TokenHub or CloudBase invoices and has not measured cache-hit rates; every figure below comes from the Tencent Cloud notice (announce detail 2542).
Related DigIntoAI notes: Tencent Hy4 Hits TokenHub and OpenRouter—Vision In, 50% Off Through Year-End, Alibaba Cloud Model Studio Token Plan Softens Night Credits for Qwen and DeepSeek, OpenAI API Pricing (November 2026), and Tencent Yuanbao Search WSA Pricing (October 2026).
What changes on Nov 6 (cache-hit input only)
The notice title names both TokenHub and CloudBase. The opening sentence says TokenHub will adjust cache-hit input for Step 5 Preview; the detail table is the source of record:
| Region | Billing item | Before | After |
|---|---|---|---|
| Guangzhou | Cache-hit input | 0.35 RMB / M tokens | 0.70 RMB / M tokens |
| Singapore | Cache-hit input | 0.36 RMB / M tokens | 0.72 RMB / M tokens |
Both rows are exactly 2× the prior cache-hit input line. Nothing else in the adjustment table is listed as changing.
What does not change (per the notice)
Under “items not adjusted,” the notice says cache-miss input and output prices remain unchanged. It does not print the current miss-input or output numbers in this announce—so DigIntoAI is not inventing those list prices here. For full TokenHub model pricing tables, the notice points to the TokenHub model-price document; the rate figures below still follow announce 2542, not a re-scraped price grid.
Why a doubled cache-hit line is not “the whole bill doubles”
Token usage bills usually mix several lines: cache-hit input, cache-miss input, and output (plus any other product fees outside this notice). Only the cache-hit input unit price is scheduled to double.
A simple structure sketch (illustrative—not DigIntoAI measurements):
- If a workload’s tokens are mostly cache hits, the cache-hit line is a large share of token spend, so that line’s 2× move can dominate the token subtotal.
- If hits are rare, most input sits on the unchanged miss line, and output is also unchanged—so the invoice can stay close to today’s level even after Nov 6.
- Mixed hit rates land in between: the bill moves with the share of tokens billed as cache-hit input, not with a blanket “everything ×2” rule.
The notice itself does not publish a hit-rate assumption. DigIntoAI is not claiming any production hit rate, and is not saying any specific account’s total will double.
TokenHub / CloudBase channel only
The notice frames the scope as follows:
- Applies to: Tencent Cloud TokenHub and CloudBase channels for model id
step-5-preview, cache-hit input only, from the BJT effective time above. - Does not extend to: a StepFun (阶跃星辰) website / all-channel / other-cloud synchronized price hike. DigIntoAI has not verified a matching StepFun all-channel announcement for this change, and does not treat the Tencent Cloud notice as a global Step 5 price rewrite.
If a caller disagrees with the fee change, the notice says the service can be stopped.
Timing: still a preview when this was written
| Clock | Meaning |
|---|---|
| Notice page / check window | Around 2026-11-04 Asia/Shanghai (page footer text 2026-11-4) |
| Effective time | 2026-11-06 00:00:00 Asia/Shanghai (Beijing time) |
The notice’s publication day is not the price-effective day. Until 2026-11-06 00:00:00 BJT, describing the rates as already raised would be wrong.
What DigIntoAI has not done
- No TokenHub / CloudBase invoice reconciliation for
step-5-preview - No cache-hit-rate measurement on live traffic
- No claim that miss-input or output list prices changed
- No claim of a StepFun all-channel hike
- No claim that every total bill must double on Nov 6
Product docs and notices can change; the Tencent Cloud announce and TokenHub price document take precedence over the figures here.
Quick facts
- Who / channel. Tencent Cloud TokenHub & CloudBase (not framed here as StepFun all-channel).
- Model. Step 5 Preview — id
step-5-preview. - When. Effective 2026-11-06 00:00:00 Asia/Shanghai; still a preview as of 2026-11-04 writing.
- What rises. Cache-hit input only — Guangzhou 0.35→0.70, Singapore 0.36→0.72 RMB/M tokens.
- What stays. Cache-miss input and output unchanged (numbers not listed in this notice).
- Bill angle. Cache-hit 2× ≠ automatic total-bill 2×; depends on hit share (illustrative, not measured).
- Opt-out. Disagree with the fee change → stop using the service (per notice).
- Related. Hy4 on TokenHub, Bailian night Credits, OpenAI API pricing, Yuanbao WSA pricing.
In short
Tencent Cloud’s TokenHub / CloudBase notice schedules a cache-hit input doubling for step-5-preview at 2026-11-06 00:00:00 Asia/Shanghai (Guangzhou 0.35→0.70, Singapore 0.36→0.72 RMB/M), while miss input and output stay unchanged. Written on 2026-11-04, that clock has not struck yet. A higher cache-hit unit price is not the same sentence as “the whole invoice doubles.” DigIntoAI has not checked bills or hit rates; the figures above follow Tencent Cloud announce 2542.