# Qwen2.5 3B browser LLM configuration and benchmarks

Canonical page: https://browserllm.dev/models/qwen2-5-3b/

Provider: Alibaba. Nominal parameters: 3B. [Upstream model card](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct).

Measurements describe an exact model artifact, quantization, runtime, browser, device, workload and cache state. Performance medians exclude warmup. Tokenizers and actual output lengths differ across models. These tests do not measure intelligence or task quality; no overall model score is assigned. Missing telemetry is unavailable, never zero. CPU is dedicated Chrome process-tree capacity; Chrome GPU time is experimental driver-accounted time including rendering, not GPU utilization. Power is whole-host subsystem power, not model-only power. Browser memory is summed process footprint, not deduplicated RAM.

## Pinned runnable artifacts

- Qwen2.5-3B-Instruct-q4f16_1-MLC: q4f16_1; WebLLM 0.2.82; context limit 4096 tokens; required shader features shader-f16; revision 7690aaaa46df36b1be0fe93b9c9abac0497eff6c. [Pinned model files](https://huggingface.co/mlc-ai/Qwen2.5-3B-Instruct-q4f16_1-MLC/resolve/7690aaaa46df36b1be0fe93b9c9abac0497eff6c/).
- Qwen2.5-3B-Instruct-q4f32_1-MLC: q4f32_1; WebLLM 0.2.82; context limit 4096 tokens; required shader features none beyond WebGPU; revision dfa91e859b714acfa489a1464297080656c3460d. [Pinned model files](https://huggingface.co/mlc-ai/Qwen2.5-3B-Instruct-q4f32_1-MLC/resolve/dfa91e859b714acfa489a1464297080656c3460d/).

## Published evidence

### q4f16_1 — 3ec6fd6f-698a-453a-ac99-cdcb1f70c31d

Captured: 2026-10-18T07:35:50.295Z. WebLLM 0.2.82; Chrome 155.0.0.0; M5 Pro · 48 GB · dedicated Chrome. Protocol: 60 requests · 3 × 20 · 90 s idle. Completed measured requests: 1. Output tokens per request: 63.

- Median runtime-reported decode speed: 56.78 tok/s.
- Median first-output latency: 55.55 ms.
- Median actual response time: 1.17 s.
- Sampled peak Chrome footprint: 5.25 GB.
- [Raw published JSON report](https://browserllm.dev/reports/3ec6fd6f-698a-453a-ac99-cdcb1f70c31d.json).