xpu liveBETA
Models›Qwen3 30B A3B

Qwen3 30B A3B

Model reference ↗

Model ranking

Arena Score
1316.8224184558
Arena reference rank
249
Evaluation configuration
Qwen3-30B-A3B
Source published
2026-11-19
Votes
26130
Confidence lower
1312.1028243712
Confidence upper
1321.5420125403
Last successful check
2026-11-22T14:00:05.877Z
Last value/rank change
2026-11-22T09:30:20.053Z
Methodology
Human preference — Text / overall; checked every six hours.

Source: Arena / LMArena ↗ · CC BY 4.0. Selection and catalog matching by XPU LIVE.

Pricing 1 Model information

Provider prices 1

Lowest input price per billing unit. Conditions in Details.

ProviderPriceProvider model IDStatusContext / max outputSource
Cloudflare Workers AIqwen3-30b-a3b-fp8
Input$0.0509
Output$0.335
USD / 1M tokensUpdated 2026-10-20
@cf/qwen/qwen3-30b-a3b-fp8Catalog-listed
Conditions

Public catalog listing; account eligibility and capacity can vary. Verified component tariffs are supplied by the existing pricing engine.

Context tokens 32768Official source ↗2026-10-20

Qwen3 30B A3B is a Qwen mixture-of-experts language model with a smaller active subset per token. Cloudflare’s FP8 endpoint is represented as a deployment of this checkpoint.

Qwen3 30B A3B: suitable applications

Evaluate multilingual assistance, reasoning and tool-driven text workflows that benefit from a mixture-of-experts option between smaller and larger models.

Capabilities and technical details

The route documents reasoning and function calling, while the developer describes the underlying expert architecture. Total and active parameters are recorded separately where verified.

The model accepts text and returns text. The recorded input reference is 32,768 tokens, with any lower Cloudflare cap recorded separately. The recorded terms are Apache 2.0; the linked source explains their scope.

Limitations and selection

The active parameter label does not mean the whole checkpoint has only three billion weights. FP8 precision and endpoint context limits can differ from other hosted versions.

Compare task accuracy and complete generated-token cost with a dense Qwen model, keeping reasoning settings and input evidence the same.

Cloudflare API pricing

Deployment pricing is tied to the selected endpoint and its documented token components. Compare input, output and cached-input charges only where those components are actually quoted; an absent component is not an invented free rate.

Review the official Cloudflare endpoint reference and the Qwen model documentation. Compare the linked Cloudflare Workers AI provider profile with the xpu live model directory when selecting a deployment.

CONTEXT TOKENS32,768Source not recorded
MAX OUTPUT TOKENSNot suppliedProvider-dependent
PROVIDERS1Linked in this catalog
SOURCE CHECKED2026-10-20Sources reviewed; unknown fields labeled

Specifications

Catalog model ID
qwen/qwen3-30b-a3b
Developer model ID / revision
Not verified in cited sources
Maximum output tokens
Not verified in cited sources
Structured output
Not verified in cited sources
API parameters (source-specific)
Not verified in cited sources
Weight license / access terms
Apache 2.0Developer weight license / hosted access terms ↗
Total parameters (billions)
30Developer model card / documentation ↗
Active parameters (billions)
3Developer model card / documentation ↗
Parameter-count details
Developer-declared nominal model size; serialized tensor counts can differ. Active parameters, when supplied, are separate from total parameters.Developer model card ↗
Download size (GB)
Depends on checkpoint and precision; no verified file size
Release date
Not verified in cited sources
Notes
Cloudflare deployment limits and transport options are recorded on its offering. Weight license is separate from hosted-service conditions. Context shown here is the publisher configuration; provider caps can be lower. FP8 is a Cloudflare deployment format, not a separate canonical model. Developer model card specifies 32,768 native tokens; longer YaRN contexts require a separate configuration. Parameter sizes are developer-declared nominal labels; serialized tensor counts may differ.

Catalog entries link to their specification and availability sources. This is a curated selection, not every model each provider offers. Best prices are the lowest collected USD tariffs, not a total-cost estimate. Token offers are selected by input price, with output from the same tariff; workload mix may change the cheapest provider. Listed variants may require batch processing, off-peak hours, or specific media settings. Price coverage varies by provider; collected tariffs include their source and observation time. Arena user-preference scores are checked automatically every six hours against the latest published Text / overall dataset. Source: Arena / LMArena; CC BY 4.0. Selection and catalog matching by XPU LIVE. They apply to the shown evaluation configuration; listed prices may use different settings. Models without verified scores are not assumed to be weaker.