Quality · Overall
0–100 score · higher is better
- Color = publisher
- Memory under review
Source: LiveBench · release of Jun 25, 2026 · checked Oct 18, 2026 · 37 open models without a score in this release are not drawn (no score ≠ 0)
DeepSeek V4.1 Flash · max · 81.1 LiveBench Overall
Llama 3.2 1B Instruct needs only ~1 GB on its own, the lightest of 60 open models
Smaug Mini gives the most quality per GB: 3.66 LiveBench points
Choose your machine and every chart on this page adapts to its memory.
Intelligence × memory
LiveBench Overall versus estimated memory (log scale) · up and left is better
- Open model · named when no other is both lighter and better
Source: LiveBench · release of Jun 25, 2026 · checked Oct 18, 2026 · memory: Wizard Q4 estimate · not drawn, memory under review: DeepSeek V4.1 Flash · max, Smaug Flash, DeepSeek V4 Flash Vision · exp, and Qwen3.8 Flash Next
Memory needed
Memory each open model needs on its own (model only, before a machine's reserve) · GB, log scale · lower is better
- Memory under review
Source: Wizard estimate from published parameter counts
Licenses
45 of 60 open models use a permissive license (MIT or Apache 2.0); the others use the publisher's own terms
Source: each model's public card on Hugging Face
By task
Only text models have a comparable score so far; the other tasks show what is catalogued.
Recommender
Chat and write · pick the task in the highlights above
Top 3 · LiveBench OverallFull ranking
Compare models
Quality, memory and license side by side. Up to 3 models.
- Oct 26, 2026EmbeddingGemma vs multilingual-e5-large for embeddings
- Oct 26, 2026BGE-M3 vs Qwen3-Embedding 0.6B for local search
- Oct 26, 2026Whisper large-v3 vs large-v3-turbo for transcription
- Oct 26, 2026Wan 2.2 vs HunyuanVideo for local video
- Oct 26, 2026Molmo 7B vs MiniCPM-V 4.5 for images
- Oct 26, 2026Gemma 4 26B A4B vs Qwen3-VL 30B A3B for images
- Oct 26, 2026FLUX.2 klein 4B vs Z-Image Turbo for fast images
- Oct 26, 2026HunyuanImage 3.0 vs Qwen-Image 2512 for images
- Oct 23, 2026Stable Diffusion 3.5 Large vs FLUX.1 dev for images
- Oct 26, 2026Gemma 4 31B vs Qwen3.8 27B for local text
- Oct 26, 2026Gemma 4 26B A4B vs Qwen3 30B A3B for local chat
- Oct 26, 2026Granite 4.2 8B vs Qwen3 8B for business use
- Oct 26, 2026Ministral 3 8B vs Llama 3.1 8B for local text
- Oct 26, 2026Qwen3 Coder 30B vs Devstral Small 2 for coding
- Oct 26, 2026Phi-4 vs Mistral Small 3.2 for local text
- Oct 26, 2026gpt-oss-20b vs Qwen3 30B A3B for local chat
- Oct 26, 2026Qwen3 8B vs Llama 3.1 8B for local chat
- Oct 26, 2026FLUX.2 dev vs Qwen-Image workflows
- Oct 26, 2026Qwen 27B vs 35B A3B for coding
- Oct 26, 2026Qwen 9B vs 27B for local text
20 published comparisonsOpen comparison
Catalog updates
1,672 catalog groups from 125 publishers.
NewEmbeddingGemma vs multilingual-e5-large for embeddingsOct 26, 2026
- Oct 6, 2026 · addedEmbeddingGemma 2: multimodal embeddings in 740M parametersGoogle · MTEB (Multilingual, v2) 61.4 (publisher-reported)
- Sep 10, 2026 · repository updatedDeepSeek V4.1 Flash · max
- Sep 10, 2026 · repository updatedSmaug Flash
1,672 catalog groups · checking Sep 26, 2026Full catalog