How much does Qwen3.8 Flash cost?

Qwen3.8 Flash costs $0.15 per 1 million input tokens and $0.47 per 1 million output tokens. This pricing is set by Alibaba.

What tokenizer does Qwen3.8 Flash use?

Qwen3.8 Flash uses the cl100k_base tokenizer. This tokenizer has a vocabulary of ~100,000 tokens and is widely used across many models.

What is Qwen3.8 Flash's context window?

Qwen3.8 Flash supports a context window of 1.0M tokens (1,000,000 tokens). The maximum output length is 131K tokens.

How does Qwen3.8 Flash compare to other models?

You can compare Qwen3.8 Flash's pricing and capabilities against other LLMs in the comparison table above, or visit the full pricing page for a side-by-side view of all models.

How do I count tokens for Qwen3.8 Flash?

Use the token counter at the top of this page or visit the homepage calculator. Paste your text and it will show the exact token count and cost for Qwen3.8 Flash using the cl100k_base tokenizer.

Qwen3.8 Flash

Last verified September 10, 2026 · Alibaba pricing ↗

$0.150/1M input · $0.470/1M output · 1.0M context · Alibaba

Count Tokens

Tokens

—

Input Cost

—

Output Cost

—

Tokens

—

Input Cost

—

Output Cost

Estimate Monthly Cost

Monthly Cost Estimator

Quick:

API calls / dayAvg input tokens / callAvg output tokens / call

<$0.0001/mo

Pick a preset above or enter custom usage

Alternatives to Qwen3.8 Flash

Cheaper Alternative

GPT-6 Luna

$0.100 in · $0.500 out · OpenAI

3% cheaper with same or larger context

Premium Upgrade

GLM-5.3-Flash

$0.150 in · $0.500 out · Zhipu

Upgrade with 1049K context

Same Provider

Qwen3 Coder Next

$0.120 in · $0.800 out · Alibaba

Higher tier in Alibaba ecosystem

Pricing Details

AlibabaCORRECTED 2026-09-10: Alibaba Cloud Model Studio's own pricing page lists qwen3.8-flash (International) at $0.15 input / $0.47 output per 1M, a single tier from 0 to 1M tokens, with cache hits at 10% of input ($0.015) and explicit cache creation at 125%. The $0.16 previously stored came from Alibaba's launch blog's rounding of the yuan card and from OpenRouter; the page a card is charged against says $0.15, which puts this model on the same input price as GLM-5.3-Flash and DeepSeek V4.1 Flash. The rest of this note predates the correction and still says $0.16 where it discusses the dollar card. TWO SEPARATE RATE CARDS, not one converted: mainland China meters this model at 1 yuan per million input and 3 yuan per million output and is reported 60-70% cheaper, while $0.16/$0.47 is the international (Singapore) tier's own USD card. Do NOT describe the dollar figure as an FX conversion of the yuan one - the arithmetic refuses it. 1 yuan is about $0.14 at recent rates, not $0.16, and the yuan card's output:input ratio is exactly 3.0 against the dollar card's 2.94. So the USD rate does not drift with FX; which card you pay is a function of which endpoint you call. We could not reach a first-party Alibaba Model Studio page carrying the USD rate, hence priceUnverified; OpenRouter's Alibaba endpoint is the source. Cached input $0.016, a 10x discount off input, against Z.ai's 5x on GLM-5.3-Flash. CACHE WRITE IS $0.20, i.e. 1.25x its own fresh input rate. The write REPLACES the fresh input charge rather than adding to it, so a prefix written and never reused is worse than no cache by exactly $0.04/Mtok, and break-even is 0.28 of a read: writing then reading once costs $0.216 against $0.32 for sending the same tokens twice uncached. Any prefix used even once is ahead. Do NOT compute this as 0.20/(0.16-0.016)=1.389, which wrongly treats the write as additive. supports_implicit_caching TRUE, so prefixes match automatically - a team that writes no caching code gets Alibaba's rate and not Z.ai's. Listed on OpenRouter 2026-08-26 at 19:37 UTC, 5h38m after GLM-5.3-Flash, which is its closest comparable on price, modality, context and target workload. 125B total parameters plus a 51B N-gram embedding layer, 6B active per token, multimodal MoE, text+image+video in. Native context 262,144 extended to 1M via YaRN. REASONING IS OPTIONAL: the listing reports mandatory false with reasoning-budget support, so output volume (and therefore the larger half of the bill) is cappable, which GLM-5.3-Flash's mandatory max-effort configuration does not allow. NO ARTIFICIAL ANALYSIS ENTRY as of 2026-08-27, so there is no independent index score, token count or cost-per-task; every cross-model cost figure we publish holds GLM-5.3-Flash's measured tokens constant across both, which prices the rate cards and not the models. OpenRouter P50 telemetry: 52 output tok/s, 4.68s latency, 100% uptime, one endpoint (Alibaba Cloud International). Do not put that latency in a column beside AA's TTFT figures; different measurement. Vendor-reported benchmarks, unchecked: DeepSWE 1.1 58.7, SWE-bench Pro 62.5, CoWorkBench 73.9, GPQA Diamond 91.7, LiveCodeBench v6 91.9, AndroidWorld 84.5, LVBench 76.6. Separately, Alibaba open-weighted Qwen3.8-Flash-Next on 2026-08-24, whose native context is 262K rather than the 1M sold here; it is a sibling, not this model, and self-host-vs-API math is not like-for-like. Against Qwen3.7 Flash the upgrade is not uniformly a rise: 3.7's tiering charges $0.03/$0.13 under 32K but $0.20/$0.80 above 256K, so this flat card is 5.33x dearer on short prompts and 1.25x/1.70x CHEAPER on long ones. Estimated tokens (different tokenizer).

Input / 1M tokens

$0.15

Output / 1M tokens

$0.47

Context Window

1.0M

Max Output

131K

Full Pricing Table Token Calculator

Price History

No price change recorded since 2026-08-27, when this site started tracking this model; it launched on 2026-08-26.

Frequently Asked Questions

Common questions about Qwen3.8 Flash pricing and usage

Full Pricing Table Compare Models Cost Calculator Alibaba Token Counter