KymaKyma
Alibaba

Alibaba

Qwen 3 32B

A fast, cost-efficient 32B language model optimized for code generation, mathematical reasoning, and multilingual text tasks. Developers should reach for it when they need reliable technical performance without the latency or overhead of larger models.

Modalities

Text → Text

Input

$0.108 /1M

Output

$0.378 /1M

Cached input

$0.0108 /1M90% off

Context

33K

Speed

fast

Performance

Live production data from real requests on Kyma — not synthetic benchmarks.

Rank

#14

of 87 active models

Tokens served

37.5M

all-time

Success rate

100%

last 7 days

Median throughput41 tok/s
Total requests5,483
Platform share1.6%
Tokens · last 15 daysJul 19 → Aug 2

Top apps using this model

1kyma-dub GUI548.8K tok
2Python App71.0K tok

Public apps sending the most traffic to this model — a signal of what real workloads it fits.

Pricing

Pay per token. Cached input is billed at 10% of the input rate.

$0.108 /1M input$0.378 /1M output
Hobby10 req/day · 2K in / 500 out
~$0.12/mo
Production1,000 req/day · 2K in / 500 out
~$12.15/mo
Scale20,000 req/day · 2K in / 500 out
~$243/mo
+ Estimate your workload
1,000
2,000
500
30%

Estimated monthly cost

$10.40

$0.3467 / day on Qwen 3 32B

Same workload on:

Qwen 3.7 Flash$5.73-45%
Qwen 3.7 Max$269+2484%

Estimates use list pricing with cached input billed at the 90%-discount rate. Actual bills depend on real token counts — every response includes its exact cost.

When to use Qwen 3 32B

Updated 2026-07-31

Where this model earns its cost — and where it doesn't.

Qwen 3 32B is a text-only language model built by Alibaba. It is designed around technical workloads, handling programming, mathematics, and translation across multiple languages. The model natively supports tool use, structured outputs, and explicit reasoning steps.

On Kyma, it runs with a 32,768-token context window and an 8,192-token maximum output. It operates in the fast and cheap tiers, with a median throughput of 33 tokens per second and a 100% success rate across live traffic. Prompt caching is fully supported, billing repeated prompt prefixes at 10% of the standard input rate.

The model is strictly text-in, text-out and does not support vision or multimodal inputs. While it handles reasoning and tool calls reliably, its output limit means it is not suited for generating extremely long documents or extended codebases in a single pass. Kyma routes requests through an OpenAI-compatible endpoint with automatic failover and exact cost reporting.

Automated Code Generation

Generates and refines code snippets across multiple programming languages with tool-use support.

Mathematical Problem Solving

Breaks down complex math problems using explicit reasoning steps to reach accurate solutions.

Multilingual Text Translation

Translates technical and conversational text across supported languages while preserving context.

Interactive Chat Assistants

Powers responsive conversational agents that leverage structured outputs and automatic failover.

Not ideal for: Tasks requiring image or video understanding, or workloads that need outputs exceeding 8,192 tokens in a single response.

How it compares

Against the peers people actually weigh it against.

SpecQwen 3 32BQwen 3.7 FlashQwen 3.7 Max
Input /1M$0.108$0.0458$2.56
Output /1M$0.378$0.1987$7.676
Context33K1M1M
ToolsYesYesYes
ReasoningYesYesYes
Speedfastfastmedium

Quick start

Up and running in under two minutes.

  1. 1

    Create an API key

    Sign up and grab a key from the dashboard — $0.50 free credit, no card required.

    Get API key →
  2. 2

    Make your first request

    Drop in your key and send a chat completion — fully OpenAI-compatible.

    curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-3-32b",
    "messages": [
      {"role": "user", "content": "Explain prompt caching in one paragraph."}
    ]
  }'
  3. 3

    Stream responses

    Add "stream": true to receive tokens as they arrive.

    curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-3-32b",
    "stream": true,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

FAQ

Common questions about this model.

What is the context window of Qwen 3 32B?

Qwen 3 32B has a 33K-token context window — roughly 48 pages of text in a single request.

How much does the Qwen 3 32B API cost?

$0.108 per 1M input tokens and $0.378 per 1M output tokens, with cached input at $0.0108/1M — a 90% discount on repeated prompt prefixes. No subscription; you pay only for what you use.

Does Qwen 3 32B support function calling?

Yes — Qwen 3 32B supports tool/function calling and structured outputs (JSON mode), so it works with agent frameworks out of the box.

How do I use Qwen 3 32B?

Kyma is OpenAI-compatible: point your SDK's base URL at https://kymaapi.com/v1, use your Kyma API key, and set the model to qwen-3-32b. Signing up is free and includes $0.50 of credit — no card required.

Does this model support function calling and structured JSON outputs?

Yes, it natively supports tool use and structured outputs, making it compatible with standard OpenAI-compatible SDKs on Kyma.

How does Kyma handle request routing for this model?

Every request uses automatic failover; if a serving path degrades, Kyma reroutes the call transparently while reporting the exact model used in the X-Kyma-Model header.

Can I reduce costs for repetitive prompts?

Yes, prompt caching is enabled, billing repeated prompt prefixes at 10% of the standard input rate.

Start with $0.50 free credit — no card required.Create account →

More models by Alibaba

ModelContextInputOutput
AlibabaQwen 3.7 Flash1M$0.0458$0.1987
AlibabaQwen 3.7 Plus1M$0.4482$1.793
AlibabaQwen 3.7 Max1M$2.56$7.676
AlibabaQwen 3.6 Plus131K$0.4388$2.633
AlibabaQwen 3 Coder131K$0.297$1.35
AlibabaQwen3 Embedding 8B33K$0.0135$0.00