Skip to content

Alibaba Singapore time-banded pricing review

Editions: OSS, Cloud, Enterprise. Unless stated otherwise, everything on this page ships in OSS.

Retrieved 2026-09-19 from first-party public pages and from the operator Singapore International native catalog GET /api/v1/models. These are list-price estimates, not invoice costs.

Sources:

The pricing page notes:

Night hours: 22:00 to 08:00 (UTC+8), based on billing time; other hours are daytime hours.

That window is the Alibaba Model Studio idle/busy policy. It is not the native DeepSeek weekday UTC peak/off-peak schedule. Preloop must not substitute a native DeepSeek, Z.ai, or Moonshot tariff for an Alibaba-hosted SKU.

Missing Singapore International chat SKUs

The previous seed skipped every native time_band other than Default, so Alibaba-hosted DeepSeek Flash never entered the catalog. Flow cost estimates for qwen/deepseek-v4.1-flash (and the dated 0731/0813 siblings) were empty.

Singapore International token rates from the pricing page and the Flash model page:

Model ID Band Input USD / 1M Output USD / 1M Implicit cache
deepseek-v4.1-flash idle 0.15 0.60 0.015
deepseek-v4.1-flash busy 0.30 1.20 0.03
deepseek-v4-flash-0731 idle 0.22 0.66 unpriced
deepseek-v4-flash-0731 busy 0.44 1.32 unpriced
deepseek-v4-pro-0813 idle 0.66 1.98 unpriced
deepseek-v4-pro-0813 busy 1.32 3.96 unpriced
glm-5.3 flat 1.40 4.40 unpriced

Flash implicit cache is 10% of that band's input price. The dedicated Singapore International table lists idle $0.015 and busy $0.03 per million tokens. The context-cache Billing section states the same 10% exception for deepseek-v4.1-flash. The 2026-09-19 native catalog also publishes implicit cache rows for deepseek-v4-flash-0731 and deepseek-v4-pro-0813; those native cache rates are used instead of inventing a percentage.

glm-5.3 is on the Singapore International GLM table at $1.40 / $4.40. The native catalog also publishes an implicit cache row at $0.28 per million tokens. Uncached estimates use the published base rates.

For 10,000 input tokens and 1,000 output tokens, busy Flash is $0.0042 and idle Flash is $0.0021. With 5,000 of those input tokens billed as implicit cache in busy hours, Flash is $0.00285.

Coverage this snapshot does not claim

The seed now has 253 Singapore International identifiers from the native catalog. Nine native rows published no prices (fun-asr-mtl, fun-asr-mtl-2025-08-25, fun-asr-realtime, qwen3-livetranslate-flash, qwen3-livetranslate-flash-2025-12-01, qwen3-omni-flash-2025-12-01, qwen3-omni-flash-realtime-2025-12-01, wan2.6-i2v, wan2.6-t2v) and stay unpriced. Compatible-mode aliases that are not in the native catalog (ccai-pro, qwen-coder-plus, qwen2-7b-instruct, qwen3-s2s-flash-realtime, qwq-plus-2025-03-05, tongyi-tingwu-slp) stay unpriced rather than guessed. Live 2026-09-19 compatible-mode listing on this account returned 169 IDs; of the operator chat list, 78 of 82 have native Singapore tariffs. The remaining four are the aliases above: qwen-coder-plus is published only as a Chinese mainland legacy row ($0.502/$1.004), not an International SKU, and must not inherit qwen3-coder-plus. qwq-plus-2025-03-05 has no native or International list row (only undated qwq-plus does). qwen2-7b-instruct and ccai-pro are absent from both the native catalog and the International pricing table.

Chat-shaped token pairs (including thinking, omni no-audio, and embeddings) estimate from prompt/completion tokens. Image, TTS, ASR duration, and video resolution list prices are stored as unit rates. Mixed leftover modality rates stay in extra_rates instead of being blended.

Beijing, US Global, and other regional tables are different USD or CNY lists. They are not copied into the Singapore International seed. A native GET /api/v1/models overlay still has to return both idle and busy token rates before a time-banded SKU is estimated; a single band fails closed.

Downloaded HTML SHA-256:

  • model-pricing: ca3a4da55220277eefc47d59442267b1b982742b267cd096d520924260ec2f7c
  • deepseek-v4-1-flash: e3ed0886983184147c5a1c3f8242d3a0a47c49c0543a56cbccf9185079ec616f
  • context-cache: b286570e805db9a7310ba1a365cd691545365a428c72aeaff8df92097ce28202