Pythia models

15 Pythia models from EleutherAI on Hugging Face, from 213M to 12.0B parameters, published by EleutherAI in F16 and F32, with a 2K-token context. The smallest official model, pythia-160m, needs about 0.5 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.

Pick a size

One row per official Pythia size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.

ModelParametersNative VRAMFP8 VRAMINT4 VRAMLive GPU fit (native)Est. $/hrKV cache at 32K
pythia-160m213M0.5 GB0.2 GB0.1 GBRTX 3070$0.088/hr1.1 GB
pythia-410m506M1.1 GB0.6 GB0.3 GBRTX 3070$0.088/hr3.0 GB
pythia-1b1.1B2.4 GB1.2 GB0.6 GBRTX 3070$0.088/hr4.0 GB
pythia-1.4b-deduped1.4B6.3 GB1.6 GB0.8 GBRTX 3070$0.088/hr6.0 GB
pythia-1.4b1.5B3.4 GB1.7 GB0.8 GBRTX 3070$0.088/hr6.0 GB
pythia-2.8b2.9B6.5 GB3.3 GB1.6 GBRTX 3070$0.088/hr10.0 GB
pythia-6.9b7.0B15.6 GB7.8 GB3.9 GBV100$0.110/hr16.0 GB
pythia-12b12.0B26.8 GB13.4 GB6.7 GBV100$0.220/hr22.5 GB

VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.

Official models (14)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
pythia-160m213MF160.5 GBRTX 3070$0.088/hr
pythia-160m-deduped213MF160.5 GBRTX 3070$0.088/hr
pythia-160m-seed1213MF160.5 GBRTX 3070$0.088/hr
pythia-160m-seed2213MF160.5 GBRTX 3070$0.088/hr
pythia-160m-seed3213MF160.5 GBRTX 3070$0.088/hr
pythia-410m506MF161.1 GBRTX 3070$0.088/hr
pythia-410m-deduped506MF161.1 GBRTX 3070$0.088/hr
pythia-1b1.1BF162.4 GBRTX 3070$0.088/hr
pythia-1b-deduped1.1BF162.4 GBRTX 3070$0.088/hr
pythia-1.4b-deduped1.4BF326.3 GBRTX 3070$0.088/hr
pythia-1.4b1.5BF163.4 GBRTX 3070$0.088/hr
pythia-2.8b2.9BF166.5 GBRTX 3070$0.088/hr
pythia-6.9b7.0BF1615.6 GBV100$0.110/hr
pythia-12b12.0BF1626.8 GBV100$0.220/hr

Fine-tunes and community models (1)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
pythia410m-sft-tldr405MF321.8 GBRTX 3070$0.088/hr

VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.

More from EleutherAI

Start ComfyUI on a rented GPU.

© 2026 Aquanode. All rights reserved.

All trademarks, logos and brand names are the property of their respective owners.