Qwen: Qwen3 Coder 30B A3B Instruct

Text input Text output
Author's Description

Qwen3-Coder-30B-A3B-Instruct is a 30.5B parameter Mixture-of-Experts (MoE) model with 128 experts (8 active per forward pass), designed for advanced code generation, repository-scale understanding, and agentic tool use. Built on the Qwen3 architecture, it supports a native context length of 256K tokens (extendable to 1M with Yarn) and performs strongly in tasks involving function calls, browser use, and structured code completion. This model is optimized for instruction-following without “thinking mode”, and integrates well with OpenAI-compatible tool-use formats.

Key Specifications
Cost
$$
Context
262K
Parameters
30B
Released
Jul 31, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Top Logprobs Stop Logprobs Max Tokens Tool Choice Top P Frequency Penalty Logit Bias Seed Tools Temperature Presence Penalty
Features

This model supports the following features:

Tools
Performance Summary

Qwen3 Coder 30B A3B Instruct demonstrates a balanced performance profile, excelling in specific areas while showing room for improvement in others. It performs among the fastest models, ranking in the 67th percentile for speed across benchmarks, and offers competitive pricing, placing in the 75th percentile. Its reliability is notably high, with a 92% success rate, indicating consistent and stable operation. The model exhibits exceptional performance in "Keyword Topic Relevance Classification" with perfect accuracy, making it the most accurate and cost-effective model in this category. It also shows strong capabilities in "Coding" (89% accuracy, 71st percentile) and "Mathematics" (89% accuracy, 63rd percentile), aligning with its design for code generation and structured tasks. "Reasoning" is another strength, achieving 78% accuracy (68th percentile). However, a significant weakness is observed in "General Knowledge," where it scores only 32.2% accuracy (14th percentile). Its "Hallucinations (Baseline)" accuracy is 94%, which is average (48th percentile), suggesting occasional instances of generating unverified information. "Instruction Following" and "Email Classification" also present areas for improvement, with 51.5% and 91% accuracy respectively, placing them in lower percentiles.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.1
Completion $0.3

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Nebius
Nebius | qwen/qwen3-coder-30b-a3b-instruct 262K $0.1 / 1M tokens $0.3 / 1M tokens
Alibaba
Alibaba | qwen/qwen3-coder-30b-a3b-instruct 200K $0.75 / 1M tokens $3.75 / 1M tokens
Chutes
Chutes | qwen/qwen3-coder-30b-a3b-instruct 262K $0.06 / 1M tokens $0.25 / 1M tokens
SiliconFlow
SiliconFlow | qwen/qwen3-coder-30b-a3b-instruct 262K $0.07 / 1M tokens $0.28 / 1M tokens
Chutes
Chutes | qwen/qwen3-coder-30b-a3b-instruct 262K $0.06 / 1M tokens $0.25 / 1M tokens
DeepInfra
DeepInfra | qwen/qwen3-coder-30b-a3b-instruct 262K $0.07 / 1M tokens $0.26 / 1M tokens
Novita
Novita | qwen/qwen3-coder-30b-a3b-instruct 262K $0.07 / 1M tokens $0.27 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Strategy Free Executions Accuracy Cost Duration
Other Models by qwen