LIVE
OpenAI models are coming to Google Cloud21/10/26 · OpenAI|OpenAI details ChatGPT's new advertising infrastructure21/10/26 · OpenAI|Anthropic reportedly posted a net loss of over $80 billion in H1 202620/10/26 · Anthropic|Self-Evolving Interfaces Help a Mid-Sized Open Model Beat GPT-5.5 and Claude Opus20/10/26|Anthropic says Claude helped trace the origin of NIST's mysterious ECDSA seeds16/10/26 · Anthropic|Sweden bans ChatGPT and Claude across government offices23/10/26|SpectralCanary: A Lightweight Method to Detect Unauthorized Training Data Use22/10/26|Low-bit n-gram tables cut memory use in language models22/10/26|NVIDIA releases OpenShell, a runtime for fleets of autonomous AI agents22/10/26 · NVIDIA|Ai2 releases OlmoEarth v1.3, an upgraded geospatial embeddings model22/10/26 · Ai2|NVIDIA Introduces Minerva, an Observability Platform for AI Factories22/10/26 · NVIDIA|Google DeepMind maps the world's forest types at global scale22/10/26 · Google DeepMind|OpenAI models are coming to Google Cloud21/10/26 · OpenAI|OpenAI details ChatGPT's new advertising infrastructure21/10/26 · OpenAI|Anthropic reportedly posted a net loss of over $80 billion in H1 202620/10/26 · Anthropic|Self-Evolving Interfaces Help a Mid-Sized Open Model Beat GPT-5.5 and Claude Opus20/10/26|Anthropic says Claude helped trace the origin of NIST's mysterious ECDSA seeds16/10/26 · Anthropic|Sweden bans ChatGPT and Claude across government offices23/10/26|SpectralCanary: A Lightweight Method to Detect Unauthorized Training Data Use22/10/26|Low-bit n-gram tables cut memory use in language models22/10/26|NVIDIA releases OpenShell, a runtime for fleets of autonomous AI agents22/10/26 · NVIDIA|Ai2 releases OlmoEarth v1.3, an upgraded geospatial embeddings model22/10/26 · Ai2|NVIDIA Introduces Minerva, an Observability Platform for AI Factories22/10/26 · NVIDIA|Google DeepMind maps the world's forest types at global scale22/10/26 · Google DeepMind|
Hardware Match

Which model runs on my machine?

Pick your configuration. For each: the open-weight models that fit in memory, the quantization level, the remaining margin and the command to run them. Computed from actual weight sizes, not from promises.

Graphics cards

RTX 4060 · 8 Go
52 models
8 usable GB · Entry level
See the list
RTX 3060 · 12 Go
62 models
12 usable GB · Good VRAM per euro
See the list
RTX 4070 · 12 Go
62 models
12 usable GB
See the list
RTX 4060 Ti · 16 Go
66 models
16 usable GB
See the list
RTX 4070 Ti Super · 16 Go
66 models
16 usable GB
See the list
RTX 5080 · 16 Go
66 models
16 usable GB
See the list
RTX 3090 · 24 Go
86 models
24 usable GB · Second-hand: best GB per euro
See the list
RTX 4090 · 24 Go
86 models
24 usable GB
See the list
AMD RX 7900 XTX · 24 Go
86 models
24 usable GB · ROCm / Vulkan
See the list
RTX 5090 · 32 Go
87 models
32 usable GB
See the list
2 × RTX 4090 · 48 Go
90 models
48 usable GB · Tensor parallelism
See the list
RTX 6000 Ada · 48 Go
90 models
48 usable GB · Workstation
See the list
NVIDIA L40S · 48 Go
90 models
48 usable GB · Inference server
See the list
NVIDIA A100 · 40 Go
87 models
40 usable GB
See the list
NVIDIA A100 · 80 Go
97 models
80 usable GB
See the list
NVIDIA H100 · 80 Go
97 models
80 usable GB
See the list
4 × RTX 4090 · 96 Go
98 models
96 usable GB
See the list
2 × H100 · 160 Go
99 models
160 usable GB
See the list
8 × H100 · 640 Go
113 models
640 usable GB · Standard HGX server node, rented from a GPU cloud or bought.
See the list
8 × H200 · 1128 Go
114 models
1128 usable GB · HGX H200 node: very large models at full precision and long contexts.
See the list

Mac Apple Silicon

Mac Apple Silicon 16 Go
62 models
12 usable GB · Unified memory
See the list
Mac Apple Silicon 24 Go
72 models
18 usable GB · Unified memory
See the list
Mac Apple Silicon 32 Go
86 models
24 usable GB · Unified memory
See the list
Mac Apple Silicon 48 Go
87 models
36 usable GB · Unified memory
See the list
Mac Apple Silicon 64 Go
90 models
48 usable GB · Unified memory
See the list
Mac Apple Silicon 96 Go
94 models
72 usable GB · Unified memory
See the list
Mac Apple Silicon 128 Go
98 models
96 usable GB · Unified memory
See the list
Mac Studio 192 Go
98 models
144 usable GB · Unified memory
See the list

No GPU (CPU)

Laptop 8 Go RAM
34 models
6 usable GB · No GPU: CPU only, slow but possible
See the list
Laptop 16 Go RAM
62 models
12 usable GB · CPU only
See the list
PC 32 Go RAM (CPU)
86 models
25 usable GB · CPU only
See the list
PC 64 Go RAM (CPU)
90 models
51 usable GB · CPU only, large models are slow
See the list