Country
France
Tracked models
162
Frontier models
1
Largest training run
2.3 x 10^24 FLOP
162 tracked AI models have at least one contributing organization based in France. 1 of them is frontier models — among the largest training runs of their moment. The largest is Mixtral 8x22B, at 2.3 x 10^24 FLOP (April 2024). The most recent tracked release is from April 2026.
It hosts 14 tracked GPU clusters.
Models (80+)
Mixtral 8x22B2.3 x 10^24 FLOPChameleon-34B1.6 x 10^24 FLOPMistral Small 31.2 x 10^24 FLOPMixtral 8x7B7.7 x 10^23 FLOPMulti-Token Prediction 7B3.8 x 10^23 FLOPBLOOM-176B3.7 x 10^23 FLOPFragLlama: Next-fragment prediction for molecular design3.3 x 10^23 FLOPLightOn Mini2.4 x 10^23 FLOPMulti-Token Prediction 13B1.5 x 10^23 FLOPXLMR-XXL3.4 x 10^22 FLOPPleias-RAG-1B3 x 10^22 FLOPPleias 1.0 1.2B3 x 10^22 FLOPXGLM-7.5B2.3 x 10^22 FLOPSEER1.8 x 10^22 FLOPBLOOM-7.1B1.5 x 10^22 FLOPGenerative BST1.4 x 10^22 FLOPV-JEPA 29.1 x 10^21 FLOPDINOv27.4 x 10^21 FLOPHuBERT5.5 x 10^21 FLOPMSA Transformer5.5 x 10^21 FLOPESM1b5.1 x 10^21 FLOPReason-ModernColBERT3.7 x 10^21 FLOPGTE-ModernColBERT-v13.7 x 10^21 FLOPIncoder-6.7B3 x 10^21 FLOPPleias-RAG-350m2.7 x 10^21 FLOPPleias 1.0 350m2.7 x 10^21 FLOPKEPLER1.7 x 10^21 FLOPMACE-MP-08.8 x 10^20 FLOPRita-XLarge8.6 x 10^20 FLOPCamemBERT8.3 x 10^20 FLOPALiBi (L=3072, Lvalid = 3072)8.1 x 10^20 FLOPDD-PPO7.8 x 10^20 FLOPGemNet-OC5.4 x 10^20 FLOPESM1-670M (UR50/D)4.8 x 10^20 FLOPBig Transformer for Back-Translation4.8 x 10^20 FLOPMasked Autoencoders ViT-H4.6 x 10^20 FLOPESM1-670M (UR50/S)4.4 x 10^20 FLOPTransformer-XL + SIS3.8 x 10^20 FLOPPAGnol-XL2.6 x 10^20 FLOPViT + DINO2.1 x 10^20 FLOPContriever1.6 x 10^20 FLOPESM1-670M (UR100)1.4 x 10^20 FLOPESM1v1.4 x 10^20 FLOPPeTriBERT10^20 FLOPDeiT-B7.9 x 10^19 FLOPESM1-85M5.6 x 10^19 FLOPTransformer (Adaptive Input Embeddings) WT1034.5 x 10^19 FLOPFAIRSEQ Adaptive Inputs3.2 x 10^19 FLOPBase LM + kNN LM + Continuous Cache3.1 x 10^19 FLOPESM1-43M2.8 x 10^19 FLOPMultiBand Diffusion2.6 x 10^19 FLOPSandwich Transformer2.4 x 10^19 FLOPAll-attention network + adaptive span1.2 x 10^19 FLOPcode2seq1.2 x 10^19 FLOPNovae1.1 x 10^19 FLOPMGK 4 heads (medium)8.9 x 10^18 FLOPFeedback Transformer7.7 x 10^18 FLOPDeLighT3.8 x 10^18 FLOPShortformer3 x 10^18 FLOPMGK 8 heads (small)3 x 10^18 FLOPRetinaNet-R1012.1 x 10^18 FLOPBrain2Qwerty1.6 x 10^18 FLOPTransE1.3 x 10^18 FLOPDDIM9 x 10^17 FLOPDDPM9 x 10^17 FLOPAlphaX-18.9 x 10^17 FLOPVDCNN (on Amazon Review Full dataset)5.7 x 10^17 FLOPcode2vec3.2 x 10^17 FLOPPluribus6.6 x 10^16 FLOPFisher Vector image classifier9.1 x 10^13 FLOPMistral Medium 3.5April 2026Devstral 2 (24B)December 2025Devstral 2 (123B)December 2025Ministral 3 8BDecember 2025Ministral 3 14BDecember 2025Mistral 3 LargeDecember 2025Ministral 3 3BDecember 2025Magistral Medium 1.2September 2025Magistral Small 1.2September 2025Mistral Medium 3.1August 2025
GPU clusters (14)
Sesterce Synapse Phase 24.1K H100eSesterce H100s Phase 24.1K H100eCEA EXA1-HE Phase 32.7K H100eSesterce Synapse Phase 12K H100eCEA EXA1-HE Phase 21.9K H100eJean Zay Supercomputer Phase 41.6K H100eScaleway Nabuchodonosor1K H100eGENCI Adastra262 H100eTotalEnergies Pangea III214 H100eJean Zay Supercomputer Phase 2170 H100eJean Zay Supercomputer Phase 3131 H100eJean Zay Supercomputer Phase 165.94 H100eAtos Spartan260.54 H100eCommissariat a l'Energie Atomique Topaze60.54 H100e
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
Other countries