llmrun
Reasoning

LiveBench 2026 Browser Leaderboard

LiveBench 2026 Browser measures tasks a model completes by operating a web browser. It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.

Source: livebench15 open models ranked+33 proprietaryData through Sep 2026

Open modelsAll models

All models ranked on LiveBench 2026 Browser

Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.

#ModelScoreVRAM @ Q4
1Claude Fable 5.1 · proprietary
100.0
API only
2Claude Opus 4.8 · proprietary
96.0
API only
3Claude Fable 5 · proprietary
92.0
API only
4Gemini 4 Argon · proprietary
92.0
API only
5GPT 6 Sol · proprietary
92.0
API only
6Claude Haiku 5.5 · proprietary
88.0
API only
7GPT 5.6 Sol · proprietary
88.0
API only
8GPT 6.1 Sol · proprietary
88.0
API only
9MiMo V2.6 Pro RL · 1020B
88.0
~613 GB
10Claude Opus 5 · proprietary
84.0
API only
11Claude Sonnet 5.5 · proprietary
84.0
API only
12GLM 5.3 Flash · 321.3B
84.0
~195 GB
13GPT 5.6 Terra · proprietary
84.0
API only
14GPT 6 Astra · proprietary
84.0
API only
15Muse Spark 1.3 · proprietary
81.0
API only
16Claude Opus 5.5 · proprietary
80.0
API only
17DeepSeek V4.1 Flash · 763.2B
80.0
~458 GB
18Claude Sonnet 5 · proprietary
76.0
API only
19GPT 6 Luna · proprietary
72.0
API only
20Muse Wonder 1.0 · proprietary
72.0
API only
21Claude Sonnet 4.6 · proprietary
68.0
API only
22GPT 5.6 Luna · proprietary
68.0
API only
23DeepSeek V4 Flash Vision Exp · 304.6B
64.0
~183 GB
24Gemini 3.8 Flash · proprietary
64.0
API only
25Grok 4.6 · proprietary
64.0
API only
26Grok 4.7 · proprietary
60.0
API only
27Gemini 3.7 Flash · proprietary
56.0
API only
28GPT 5.5 · proprietary
48.0
API only
29Qwen3.8 Flash · proprietary
48.0
API only
30Grok 4.5 · proprietary
44.0
API only
31Claude Haiku 4.5 · proprietary
40.0
API only
32Muse Spark Open 816B A42B · 822.0B
36.0
~543 GB
33Qwen3.8 27B · 27.8B
32.0
~17 GB
34Qwen3.8 Max · proprietary
32.0
API only
35Muse Spark 1.2 · proprietary
29.4
API only
36Step 5 · proprietary
24.0
API only
37Inkling · 952.4B
16.0
~572 GB
38Kimi K3 · 2779.9B
16.0
~1674 GB
39MiniMax M3 · 427.0B
8.0
~257 GB
40Mistral Large 4 · proprietary
8.0
API only
41Gemini 3.1 Pro · proprietary
4.0
API only
42Gemma 4 31B IT · 31.3B
4.0
~20 GB
43NVIDIA Nemotron 3.5 Super VL 120B A12B BF16 · 124.5B
4.0
~82 GB
44Beam 1.0 · proprietary
0.0
API only
45GLM 5.3 · 753.3B
0.0
~456 GB
46NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B
0.0
~370 GB
47NVIDIA Nemotron 3.5 Lightning 30B A3B BF16 · 31.6B
0.0
~19 GB
48Smaug Flash · 304.2B
0.0
~609 GB

Score vs model size

Which models give the most quality for their size — the ones worth running locally.

100B1Tmodel size (log scale) →88.00.0DeepSeek V4.1 Flash · 763B · 80.0Muse Spark Open 816B A42B · 822B · 36.0Kimi K3 · 2.8T · 16.0Inkling · 952B · 16.0MiniMax M3 · 427B · 8.0Gemma 4 31B IT · 31B · 4.0NVIDIA Nemotron 3.5 Super VL 120B A12B BF16 · 124B · 4.0Smaug Flash · 304B · 0.0NVIDIA Nemotron 3.5 Lightning 30B A3B BF16 · 32B · 0.0NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 561B · 0.0GLM 5.3 · 753B · 0.0Qwen3.8 27B · 28B · 32.0Qwen3.8 27BDeepSeek V4 Flash Vision Exp · 305B · 64.0DeepSeek V4 Flash Vis…GLM 5.3 Flash · 321B · 84.0GLM 5.3 FlashMiMo V2.6 Pro RL · 1T · 88.0MiMo V2.6 Pro RL
Each dot is a model. Up = higher score, left = smaller (easier to run locally). The dashed line marks the efficiency frontier — the best score you can get at each size or smaller.
  • Qwen3.8 27B, 28B, score 32.0 — on the efficiency frontier (best score at its size or smaller).
  • DeepSeek V4 Flash Vision Exp, 305B, score 64.0 — on the efficiency frontier (best score at its size or smaller).
  • GLM 5.3 Flash, 321B, score 84.0 — on the efficiency frontier (best score at its size or smaller).
  • MiMo V2.6 Pro RL, 1T, score 88.0 — on the efficiency frontier (best score at its size or smaller).

LiveBench 2026 Browser: frequently asked questions

What is the best open LLM on LiveBench 2026 Browser?
MiMo V2.6 Pro RL is the top open model on LiveBench 2026 Browser, scoring 88.0. Among all models tested — including proprietary ones — it ranks #6. The top model overall is Claude Fable 5.1 (Anthropic) at 100.0.
What's the best LiveBench 2026 Browser model you can run on a 24 GB GPU?
Qwen3.8 27B is the highest-scoring open model that fits in 24 GB at 4-bit quantization (~17 GB), scoring 32.0 on LiveBench 2026 Browser.
Can open models match proprietary models on LiveBench 2026 Browser?
Not quite on LiveBench 2026 Browser: the strongest proprietary model (Claude Fable 5.1) scores 100.0, ahead of the best open model (MiMo V2.6 Pro RL) at 88.0 — but you can run the open one yourself.

Scores aggregated from livebench. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.