Skip to content
Tier B — Production
Runs in:FranceMade in:China
OVH AI Endpoints (GRA)

Qwen3-Coder-30B-A3B-Instruct

Tier B — Production

Tokonomix Editorial Team·Reviewed by Mes Kalkan··
Section 01

Speed analysis

Latency measured across all benchmark runs. P50 (median) and P95 (95th percentile) give a realistic picture of response speed under normal and peak load.

P50 latency (median)P95 latency100 runs
60792015780236403150008-2109-14ms
Section 02

Quality scores

How this model compares to the rest of the field on each prompt category, from a pairwise fit over the same prompts. The raw judge score sits underneath each number.

55%
Coding
judge mean 92
36%
Creative
judge mean 78
57%
Factual
judge mean 77
53%
Multilingual
judge mean 98
64%
Reasoning
judge mean 93

Win rate per category: how often this model beats a field-average model on a prompt from that category. 50% is average, not a failing grade. It is not a percentage of correct answers.

Section 03

Pricing

What you pay per million tokens when you use this model on Tokonomix, plus an estimate for a typical conversation.

💰
API rates — Qwen3-Coder-30B-A3B-Instruct
$0.1900 per 1M input tokens
$0.6800 per 1M output tokens
≈ $0.0003 per typical conversation (800 tokens)
Input vs output price (per 1M tokens)
per 1M input tokens$0.1900
per 1M output tokens$0.6800
Section 04

Tokens per second

Throughput in tokens per second, derived from measured P50 latency. Higher is better; fluctuations track provider-side load.

Throughput (tokens / s)2439 / avg 1626
328420

Estimated from P50 latency × 200 output tokens — the absolute number depends on this assumption; the trend is what matters.

Section 05

Capabilities

ownedBy: Qwen
Section 06

Availability

Availability

No measurements yet

We haven't recorded enough API calls to show availability stats for this model. Data appears once the model starts receiving live traffic.

Section 07

Tokonomix benchmark verdicts

⚖️
Endorsed by 2 judges
Independent LLM judges evaluated this model on our weekly intelligence tests
cohere/command-a100/100 · 1 runs
1 correct0 partial0 wrong100% accuracy
claude-sonnet-4-589/100 · 82 runs
68 correct4 partial10 wrong83% accuracy
2026-09-13

Quality falls 9.5 points to 75.2 with near-doubled latency

Qwen3-Coder-30B-A3B-Instruct demonstrates significant performance degradation in this benchmark window. Overall quality dropped from 84.7 to 75.2, representing a 9.5 point decline that follows a previous 6.5 point decrease. This marks a concerning downward trend across consecutive windows. Latency deteriorated substantially, with p50 response times increasing 93% from 2009ms to 3886ms, nearly doubling the wait time for users. Category performance shows mixed results with sharp variations. Reasoning achieved a perfect 100 score, indicating strong logical capabilities. However, coding performance plummeted from 94 to 69, a 25 point drop that undermines the model's core positioning as a coding specialist. Factual accuracy scored 57, though this represents a new category without direct comparison. The previous window's multilingual and creative categories were not tested in the current period. The combination of declining quality metrics and significantly increased latency suggests potential infrastructure or model configuration issues. Users should expect notably slower responses and reduced coding performance compared to the previous benchmark period. The perfect reasoning score provides limited consolation given the substantial regression in the model's primary coding capabilities.

Quality

75.2

Latency p50

3,886 ms

Test runs

5

Quality dropped 9.5 points Latency increased 93% Coding score fell from 94 to 69 Perfect reasoning score achieved
Last automated test
Sep 14, 2026 · 20:02 UTC · Speed benchmark
P50 latency
82 ms
P95 latency
84 ms
Errors
0 / 6 runs
Last reviewed by Tokonomix Team·September 14, 2026