Qwen3.8-Flash
Announced 2026-08-26; QwenCloud list $0.16 / $0.47; this id is not in our catalog yet
Alibaba’s Qwen team posted Qwen3.8-Flash on 2026-08-26 (preview weights often branded Qwen3.8-Flash-Next). Public hosted list price is $0.16 input / $0.47 output per 1M tokens, with a 1M context pitch. This page tracks official API, weights, and Bailian RMB — it does not write this id as already present on QCode’s /models list.
Four facts you can quote
Public QwenCloud USD list
Launch write-ups and pricing pages put the production SKU at $0.16 in / $0.47 out per million. That is Alibaba’s list, not a reseller invoice.
Announce day
@Alibaba_Qwen that day; Reuters / Bloomberg same day. Weights on Hugging Face / ModelScope. Whether the production API is live for everyone is inconsistent across write-ups — this page treats rollout as unfinished.
Pitched context
Reports: native about 262,144 tokens, YaRN to 1M. Trust the model card. Do not treat a recap window as a number you already measured on an aggregator.
Keep it apart from Qwen3.7
QCode /models currently matches Tongyi ids in the qwen3.7-max / qwen3.7-plus generation. qwen3.8-flash and qwen3.8-max are absent from the 30-day usage table. Need Tongyi today? Use 3.7 already on the list.
Where Flash sits in the 3.8 family
Qwen3.8-Max shipped around 2026-08-03 as the flagship (public $2 / $6). Flash is the cheap tier of the same generation; vendor copy frames it as near-flagship at roughly one-twelfth the price. Preview weights (Flash-Next) and the hosted production SKU (Qwen3.8-Flash) share a name family — do not treat them as one checkpoint. Downloadable weights ≠ your account is billed on a hosted id.
Do not merge RMB and USD into one row
Alibaba Cloud Bailian notice 2026-08-26 23:18: from 2026-08-27 12:00 Beijing time, Qwen3.8-Flash input ￥1.00 → ￥0.80, output ￥3.00 → ￥2.70 per million tokens. International pages quote $0.16 / $0.47. Keep unit and channel in the same sentence. Bailian RMB is not QwenCloud USD, and neither is a QCode price.
Timeline
Qwen3.8-Max flagship, public hosted list about $2 / $6, 1M context. That page also said it was not in the QCode catalog then.
Qwen3.8-Flash / Flash-Next announced. USD list $0.16 / $0.47. Weights and tech report on GitHub / HF / ModelScope.
Bailian cuts RMB unit prices to ￥0.80 / ￥2.70. As of 2026-08-30 this page still splits QwenCloud USD and Bailian RMB.
Confirmed vs rumor
Confirmed
Announce day 2026-08-26, QwenCloud USD $0.16 / $0.47, Bailian RMB cut on 08-27, Max vs Flash price story, public weights — primary or official notices. No qwen3.8* id in this site’s usage table on 2026-08-30, also checked.
Rumor / misread
“Flash must run on a laptop” — parameter counts disagree across recaps; trust the card; the name is not a small-model guarantee. “Live on every aggregator” — some say API coming soon, others say Bailian already bills. This page does not pick a global live. “QCode already lists 3.8-Flash” — catalog and usage table still lack the id.
Tongyi you can call today vs 3.8-Flash still in tracking
Need to run today: Qwen3.7 already on the list
qwen3.7-max / qwen3.7-plus show up in the usage table. If the job cannot wait for 3.8, use those. Price and window follow the live /models list.
3.8-Flash: follow official channels; do not hard-code yet
Official API, Bailian, and HF weights are three landings. Before production hard-codes qwen3.8-flash, confirm that console actually returns the id. Until this catalog updates, do not treat it as a provisioned route.
How to track (official channels)
Watch @Alibaba_Qwen, the qwen.ai blog, Bailian notices, and the model card — not only aggregator mirrors. USD gets a dollar sign, RMB a ￥, channel in the same sentence. Need Tongyi immediately? Use 3.7 ids already on this catalog. This page does not describe forging an unprovisioned id into a live route.
This site does not describe Qwen3.8-Flash as callable here
QCode’s /models list and 30-day usage table still have no qwen3.8-flash. Tongyi ids that currently match are qwen3.7-max and qwen3.7-plus. The 3.8-Flash story lives on official cloud and open weights, not in this catalog. Until the id shows up in usage records, this page stays a tracker and does not change that sentence.
FAQ
What is the public list price for Qwen3.8-Flash?
International pages use $0.16 in / $0.47 out per million (QwenCloud). Bailian RMB after 2026-08-27 12:00 Beijing time is ￥0.80 / ￥2.70. Trust the console you actually pay.
How far from Qwen3.8-Max?
Max public list is about $2 / $6. Several write-ups frame Flash at roughly one-twelfth. That is vendor tier copy, not a score we measured.
Are Qwen3.8-Flash-Next and Flash the same id?
Next is usually the architecture-preview / weights name; Flash is the hosted production SKU. Do not paste an HF repo name into a random vendor’s model field.
Can I call qwen3.8-flash on QCode today?
Do not plan as if it were already on the list. The 2026-08-30 usage table has no qwen3.8*. For Tongyi, pick qwen3.7-max or qwen3.7-plus.
Is 1M context on by default?
The pitch is 1M; native length is reported near 262K plus YaRN. Default-on and extra billing follow the official model note. This page does not invent a hidden flag.
Can I run it locally?
Weights are on HF / ModelScope. VRAM and active parameters follow the card. The word Flash does not guarantee a consumer GPU will run the full hosted spec comfortably.
Sources
@Alibaba_Qwen announce post 2026-08-26; launch coverage of QwenCloud $0.16 / $0.47; Alibaba Cloud Bailian notice 2026-08-26 23:18 (￥0.80 / ￥2.70 from 2026-08-27 12:00 Beijing time). Max background: the on-site Qwen3.8-Max tracker. Catalog conclusion: 2026-08-30 usage table, no qwen3.8*. Fetched 2026-08-30.
Tongyi 3.8-Flash is still landing on official channels; we track it here
The catalog has no qwen3.8-flash. Need Tongyi now? Use 3.7 already on the list. Platform billing is official price × service fee.
Related
Qwen3.8-Max tracker
Same-generation flagship: public $2/$6, also not in this catalog.
Qwen3.8-27B
A downloadable 3.8 small tier — not the same id as the hosted Flash SKU.
Qwen3.7-Plus
One of the Tongyi tiers that actually matches today’s catalog.
Reading of Qwen / Alibaba Cloud public notices and press recaps, not an Alibaba statement. Price and entitlement follow the console you use. QCode does not describe an id missing from catalog and usage as provisioned.