.published-loading { display: none; }

Browser LLM leaderboard

Compare browser LLMs by generation speed, first-output latency, response time, memory and resource use.

Results describe the recorded device, browser and workload. Answer accuracy is not tested; output lengths differ across models.

Compare models

10 measured configurations · 60 requests · 3 × 20 · 90 s idle

Published browser LLM measurements, excluding warmup
Model	Generation speed (tok/s)	First-output latency (ms)	Response time (s)	Peak browser memory (GB)	Completed requests
Llama 3.2 1Bq4f16_1 · WebLLM 0.2.82	163.6	26.3	0.42	3.82	1
Qwen2.5 0.5Bq4f16_1 · WebLLM 0.2.82	113.8	40.1	0.60	2.70	1
Qwen2.5 1.5Bq4f16_1 · WebLLM 0.2.82	79.1	40.0	0.84	4.05	1
Qwen2.5 3Bq4f16_1 · WebLLM 0.2.82	56.8	55.6	1.17	5.25	1
Mistral 7B Instruct v0.2q4f16_1 · WebLLM 0.2.82	51.9	223.0	1.44	7.08	1
Mistral 7B Instruct v0.3q4f16_1 · WebLLM 0.2.82	51.1	165.0	1.41	7.12	1
Qwen2.5 7Bq4f16_1 · WebLLM 0.2.82	50.4	94.7	1.35	9.08	1
Qwen3 1.7Bq4f16_1 · WebLLM 0.2.82	47.2	55.8	1.30	4.78	1
Qwen3 0.6Bq4f16_1 · WebLLM 0.2.82	38.7	52.8	1.35	3.73	1
Qwen3 4Bq4f16_1 · WebLLM 0.2.82	29.6	102.6	2.10	6.63	1

Request metrics are medians. Peak memory is a sampled sum of Chrome process footprints, not deduplicated RAM. Missing readings stay unavailable. Measurement definitions