LONG AI
Home › Blog › Step 5 Preview Cache-Hit Input Price Dou

Step 5 Preview Cache-Hit Input Price Doubles Nov 6

Category:资讯 · 5 min read · 2026-11-05

On Nov 4, Tencent Cloud posted a repricing notice: Step 5 Preview (model ID step-5-preview) doubles its "cache-hit input" price, effective Beijing time 2026-11-06 00:00:00. See Tencent Cloud announcement detail/2542, published 2026-11-04 15:06:58. Only this one billing item rises; uncached input and output prices are explicitly "unchanged." No retirement, no rename - the call method stays the same.

Step 5 Preview cache-hit input price doubles Nov 6: Guangzhou 0.35 to 0.70, Singapore 0.36 to 0.72 yuan/M tokens, still on sale on our plaza
Fig 1 - The gist: only the cache-hit-input tier moves, Guangzhou and Singapore both double, effective Nov 6 00:00. Data from Tencent Cloud announcement detail/2542, self-made.

1. Official adjustment detail

RegionBilling itemBeforeAfterChange
GuangzhouCache-hit input0.35 yuan / M tokens0.70 yuan / M tokens+100%
SingaporeCache-hit input0.36 yuan / M tokens0.72 yuan / M tokens+100%
All regionsUncached inputunchanged (no figure listed)same
All regionsOutputunchanged (no figure listed)same

The announcement only gives two operational tips: review call volume and cost expectations ahead of time, and set up usage monitoring; discontinue use if you disagree. No transition period, no named replacement - a pure repricing, not a retirement.

2. Who actually pays more

A cache hit means multiple requests reuse the same prefix (long system prompts, fixed knowledge snippets, multi-turn history) - that segment skips recomputation and bills at a lower rate. It's exactly this tier going up. High-hit-ratio scenarios are the most sensitive: long system prompts with batch repetitive tasks, RAG fixed documents, long multi-turn sessions. One-off short requests see near-zero change; output-heavy workloads are unaffected too.

A ruler you can compute yourself: new cost = cache-hit volume (M tokens) × 0.35 yuan (Guangzhou) or 0.36 (Singapore). Example: 1,000 M tokens of hits a month, Guangzhou, roughly ¥350 more - reconcilable against the console's usage detail.

3. Our plaza test: in stock, but no separate "cache-hit" tier

At 15:55 on 2026-11-05 we pulled our /api/v1/model-plaza API and cross-checked the page:

Plaza search for step-5-preview: 6 sellers, lowest 1.6 off, buyer price 1.16 in / 3.30 out per M tokens, official base 7.00 / 20.00
Fig 2 - Our plaza (fetched 2026-11-05 15:55): step-5-preview in stock, 6 sellers, lowest 1.6 off, no separate cache-hit price in the quote.

Math you can re-verify (same method as our token billing article): official base is USD $1.04556 / $2.9873 per M tokens, converted at our uniform rate 6.695 to ¥7.00 / ¥20.00; lowest tier buyer_percent_off = 83.5 (page shows 1.6 off). ¥7.00 × 16.5% ≈ ¥1.16, ¥20.00 × 16.5% = ¥3.30, matching the page.

Self-made comparison: left Tencent official cache-hit price 0.35 to 0.70 / 0.36 to 0.72, right our plaza official base 7.00/20.00 and lowest 1.16/3.30, middle the Nov 4 to Nov 6 timeline
Fig 3 - Self-made comparison: left Tencent official before/after, right our plaza price and in-stock status, middle the "Nov 4 announced to Nov 6 effective" timeline, bottom the math.

4. What it means for ordinary users

  1. Direct to Tencent Cloud TokenHub: auto-bills at the new price from Nov 6 00:00; no code change, no model rename - the cache-hit portion doubles.
  2. Via our plaza: page price and model name are both unchanged today. We can't confirm whether our sellers' upstream passes through Tencent TokenHub billing, so we don't predict or promise whether our price will follow; come back after Nov 6 to re-check the quote - plaza price forms in real time from seller bids.
  3. What not to do: the official side announced no retirement, no context or rate-limit change, and no migration requirement - no need to rush a model switch.

5. Our advice: three steps

  1. Measure your hit ratio today: pull the last 7 days of usage detail from the Tencent console, note the "cache-hit input tokens," and use the formula above to see how much more you'll pay monthly.
  2. High ratio - save before switching models: consolidate repetitive long system prompts into a fixed prefix, merge sessions, clean up redundant documents resent every turn. With the hit price doubled, cutting useless hits is worth as much as keeping useful ones.
  3. Re-check two places after Nov 6: whether Tencent's hit price is already 0.70 / 0.72, and our plaza's in-stock status and lowest quote for step-5-preview.

Scope note

Official info from Tencent Cloud announcement cloud.tencent.com/announce/detail/2542 (published 2026-11-04, effective 2026-11-06 00:00:00), figures copied verbatim; plaza price and in-stock status from our /api/v1/model-plaza API and page, fetched 2026-11-05 15:55. Tencent's list price and the seller-bid plaza price are two separate systems - you can't subtract them to say "how much cheaper"; this article reports only price and verifiable availability, not quality or speed. Plaza price moves with supply in real time; the final price follows the model plaza page. Images are our screenshots or self-made.

相关文章

GPT-6.2 Cuts Prices 20%: What the AI Price War Means

GPT-6.2 cuts list prices 20%, yet platform discounts run deeper: gpt-6-astra trades at under 1% of list. What this week's AI price war means for buyer

What Is Prompt Caching? How Cache Tokens Cut API Bills

Prompt caching stores a repeated prompt prefix and bills cache reads at 5-12% of normal input. Real prices, hit conditions, and how to cut your API bi

All 12 Claude Models Compared: Price and Best Use

Twelve Claude versions are on sale. The newest opus-5-5 costs $8.91 a month; the oldest opus-4-6 costs $49.50. Full table and a pick for each workload

Which AI Model Writes Code Best? Speed and Price Tested

gpt-5.6-terra and deepseek-v4.1-flash win on price and reliability; gemini-3.8-flash is fastest. Effective prices, measured speed, and a three-tier pi

One API Key, Every Model: How to Integrate a Multi-Model Marketplace

Point your base URL at the platform and one key covers all 138 on-sale models across 51 vendors, in both OpenAI and Anthropic request formats. Copy-pa