Research Releases
Neuronpedia works with research groups to put their work online, so that other researchers can use it right away. Each release has its own page, and most have interactive tools.
Find all releases under Releases in the top menu. Read about new releases on the Neuronpedia blog.
Tools From Research Releases
These releases have their own sections in these docs:
| Release | Authors | Docs |
|---|---|---|
| Circuit Tracing | Anthropic, and many other groups | Circuit Tracing |
| HeadVis | Anthropic (Luger, Kamath, et al.) | Attention Heads |
| Jacobian Lens | Anthropic (Gurnee et al.) | Jacobian Lens |
| Natural Language Autoencoders | Anthropic (Fraser-Taliente, Kantamneni, Ong, et al.) | NLAs |
| Assistant Axis | Anthropic (Lu et al.) | Assistant Axis |
| Temporal Feature Analysis | Lubana, Rager, Hindupur, et al. | Temporal Feature Analysis |
| Weight-Sparse Transformers | OpenAI (Gao et al.) | Weight-Sparse Transformers |
| SAEBench | Karvonen, Rager, et al. | neuronpedia.org/sae-bench - Compare evaluation results for many SAEs. See the writeup. |
SAE and Transcoder Releases
Some of the large SAE and transcoder releases:
| Release | Models | Link |
|---|---|---|
| Gemma Scope 2 (Google DeepMind) | Gemma 3 | neuronpedia.org/gemma-scope-2 |
| Gemma Scope (Google DeepMind) | Gemma 2 | neuronpedia.org/gemma-scope |
| Llama Scope (OpenMOSS) | Llama 3.1 8B | neuronpedia.org/llama-scope |
| Goodfire Llama 3.3 70B SAE | Llama 3.3 70B Instruct | neuronpedia.org/llama3.3-70b-it-gf |
| gpt-oss SAEs (Arditi) | gpt-oss-20b | neuronpedia.org/gpt-oss-sae |
| Misaligned Persona (Arditi and Chen) | Llama 3.1 8B, Qwen 2.5 7B | neuronpedia.org/misaligned-persona |
Add Your Release
Do you have a paper with SAEs, graphs, lenses, or a new method? We can help you make a release on Neuronpedia. Email us, or see Upload Your SAEs.