llmrun
Coding

LiveBench 2026 Design Leaderboard

LiveBench 2026 Design measures layout-design tasks (LiveBench's design_layout task). 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 Design

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

#ModelScoreVRAM @ Q4
1Claude Opus 5.5 · proprietary
86.9
API only
2Claude Sonnet 5.5 · proprietary
85.0
API only
3Claude Fable 5 · proprietary
84.7
API only
4Claude Fable 5.1 · proprietary
83.4
API only
5Claude Opus 5 · proprietary
83.0
API only
6GPT 6 Luna · proprietary
82.5
API only
7Claude Haiku 5.5 · proprietary
82.2
API only
8GPT 6 Sol · proprietary
82.2
API only
9GPT 6.1 Sol · proprietary
82.1
API only
10Claude Sonnet 5 · proprietary
82.0
API only
11GPT 6 Astra · proprietary
82.0
API only
12GPT 5.6 Sol · proprietary
81.6
API only
13Grok 4.6 · proprietary
79.8
API only
14Claude Opus 4.8 · proprietary
78.8
API only
15Qwen3.8 Max · proprietary
78.6
API only
16MiMo V2.6 Pro RL · 1020B
78.5
~613 GB
17Grok 4.5 · proprietary
78.1
API only
18Grok 4.7 · proprietary
78.0
API only
19GLM 5.3 · 753.3B
77.3
~456 GB
20GLM 5.3 Flash · 321.3B
77.2
~195 GB
21Kimi K3 · 2779.9B
77.2
~1674 GB
22DeepSeek V4.1 Flash · 763.2B
77.0
~458 GB
23Step 5 · proprietary
76.9
API only
24Smaug Flash · 304.2B
76.2
~609 GB
25GPT 5.6 Luna · proprietary
76.0
API only
26GPT 5.6 Terra · proprietary
76.0
API only
27Qwen3.8 27B · 27.8B
75.9
~17 GB
28Muse Wonder 1.0 · proprietary
75.1
API only
29Muse Spark 1.3 · proprietary
74.8
API only
30Gemini 4 Argon · proprietary
73.7
API only
31Gemini 3.7 Flash · proprietary
73.2
API only
32Qwen3.8 Flash · proprietary
72.8
API only
33Gemini 3.8 Flash · proprietary
72.7
API only
34Claude Sonnet 4.6 · proprietary
71.8
API only
35DeepSeek V4 Flash Vision Exp · 304.6B
70.7
~183 GB
36GPT 5.5 · proprietary
70.3
API only
37Beam 1.0 · proprietary
67.9
API only
38Muse Spark 1.2 · proprietary
67.1
API only
39MiniMax M3 · 427.0B
66.2
~257 GB
40Mistral Large 4 · proprietary
65.6
API only
41Gemini 3.1 Pro · proprietary
61.6
API only
42NVIDIA Nemotron 3.5 Super VL 120B A12B BF16 · 124.5B
60.7
~82 GB
43Muse Spark Open 816B A42B · 822.0B
60.3
~543 GB
44Claude Haiku 4.5 · proprietary
59.9
API only
45NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B
58.6
~370 GB
46Gemma 4 31B IT · 31.3B
58.4
~20 GB
47Inkling · 952.4B
58.3
~572 GB
48NVIDIA Nemotron 3.5 Lightning 30B A3B BF16 · 31.6B
43.9
~19 GB

Score vs model size

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

100B1Tmodel size (log scale) →78.543.9Kimi K3 · 2.8T · 77.2DeepSeek V4.1 Flash · 763B · 77.0DeepSeek V4 Flash Vision Exp · 305B · 70.7MiniMax M3 · 427B · 66.2NVIDIA Nemotron 3.5 Super VL 120B A12B BF16 · 124B · 60.7Muse Spark Open 816B A42B · 822B · 60.3NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 561B · 58.6Gemma 4 31B IT · 31B · 58.4Inkling · 952B · 58.3NVIDIA Nemotron 3.5 Lightning 30B A3B BF16 · 32B · 43.9Qwen3.8 27B · 28B · 75.9Qwen3.8 27BSmaug Flash · 304B · 76.2Smaug FlashGLM 5.3 Flash · 321B · 77.2GLM 5.3 · 753B · 77.3GLM 5.3MiMo V2.6 Pro RL · 1T · 78.5MiMo 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 75.9 — on the efficiency frontier (best score at its size or smaller).
  • Smaug Flash, 304B, score 76.2 — on the efficiency frontier (best score at its size or smaller).
  • GLM 5.3 Flash, 321B, score 77.2 — on the efficiency frontier (best score at its size or smaller).
  • GLM 5.3, 753B, score 77.3 — on the efficiency frontier (best score at its size or smaller).
  • MiMo V2.6 Pro RL, 1T, score 78.5 — on the efficiency frontier (best score at its size or smaller).

LiveBench 2026 Design: frequently asked questions

What is the best open LLM on LiveBench 2026 Design?
MiMo V2.6 Pro RL is the top open model on LiveBench 2026 Design, scoring 78.5. Among all models tested — including proprietary ones — it ranks #16. The top model overall is Claude Opus 5.5 (Anthropic) at 86.9.
What's the best LiveBench 2026 Design 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 75.9 on LiveBench 2026 Design.
Can open models match proprietary models on LiveBench 2026 Design?
Not quite on LiveBench 2026 Design: the strongest proprietary model (Claude Opus 5.5) scores 86.9, ahead of the best open model (MiMo V2.6 Pro RL) at 78.5 — 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.