Open Weight Models
Home›Compare
Decision & comparison layer

Compare models by constraints, not hype.

Source-first decision pages connect the 64-model registry to hardware, workload, architecture, licensing and European deployment choices.

19 comparison & decision pages64-model registry underneathNo universal leaderboardUpdated 1 Oct 2026
How to use this section

Start with the constraint that can eliminate a model: modality, available VRAM, workload, license, deployment region or architecture. Then benchmark a short list on the exact task.

Constraint-firstFilter before benchmarking.
Source-firstFacts link to publisher sources.
Deployment-firstMemory, runtime and data flow stay visible.
Model vs model

Qwen3-32B vs gpt-oss-20b

Dense vs sparse reasoning models, both Apache 2.0.

Open comparison →Model vs model

Qwen3-32B vs Gemma 3 27B

Text reasoning versus multimodal input and different license models.

Open comparison →Coding comparison

Qwen3-Coder-30B-A3B vs Devstral Small 2

Two Apache-licensed coding-agent candidates.

Open comparison →Large reasoning comparison

DeepSeek-R1 vs Qwen3-235B-A22B

Large MoE reasoning systems with different total/active footprints.

Open comparison →Hardware shortlist

Open-Weight Models for 24 GB VRAM

Local/workstation candidates with quantization headroom.

Open comparison →Hardware shortlist

Open-Weight Models for 48 GB VRAM

20B–32B-class deployment candidates.

Open comparison →License shortlist

Apache 2.0 Open-Weight Models for Commercial Deployment

Permissively licensed candidates across size classes.

Open comparison →Workload shortlist

Open-Weight Coding Models — A Deployment Shortlist

Coding and software-engineering candidates.

Open comparison →Europe shortlist

Open-Weight Models from European Union Providers

EU-provider models with residency caveats.

Open comparison →Architecture decision

Dense vs MoE Open-Weight Models

Total vs active parameters and serving trade-offs.

Open comparison →Hardware shortlist

Open-Weight Models for 8 GB VRAM

A conservative shortlist for 8 GB VRAM, emphasizing compact checkpoints, quantization, context headroom and realistic local inference.

Open comparison →Hardware shortlist

Open-Weight Models for 16 GB VRAM

A practical 16 GB VRAM shortlist spanning 7B–14B-class open-weight models with quantization, context and runtime caveats.

Open comparison →Workload shortlist

Local Multimodal Open-Weight Models

Compare locally deployable open-weight vision-language and multimodal models across compact, workstation and larger hardware classes.

Open comparison →Workload shortlist

Open-Weight Models for RAG

A source-first RAG shortlist focused on context handling, instruction following, structured output, deployment footprint and private data paths.

Open comparison →Workload shortlist

Open-Weight Reasoning Models

Compare open-weight reasoning models across local, workstation and large-server classes without collapsing them into a universal leaderboard.

Open comparison →Capability shortlist

Long-Context Open-Weight Models

Compare long-context open-weight models while separating advertised maximum context from usable quality, KV-cache memory and serving latency.

Open comparison →Europe deployment

Open-Weight Models for EU Self-Hosting

A deployment shortlist for EU/EEA self-hosting that keeps model provider origin separate from inference location, data residency and regulatory role.

Open comparison →Family comparison

Qwen vs Mistral Open-Weight Models

Compare Qwen and Mistral open-weight model families by size range, multimodality, coding, reasoning, licenses, provider origin and deployment classes.

Open comparison →Family comparison

Qwen vs Llama Open-Weight Models

Compare Qwen and Llama open-weight families by licensing, size, multimodality, context, hardware class and self-hosted deployment.

Open comparison →