Top 50
The most expensive AI models ever trained
Ranked by the estimated cost of the compute consumed by the training run, in 2023 dollars.
Grok 4 leads with $388M, ahead of GPT-4.5 ($366M) and Grok 3 ($218M).
These figures cover the compute for the final training run only. They exclude the failed runs, the research staff, the data work and the inference infrastructure — in practice, the majority of what a frontier model actually costs its developer.
| # | Name | Organization | Published | Training cost |
|---|---|---|---|---|
| 01 | Grok 4 | xAI | Jul 9, 2025 | $388M |
| 02 | GPT-4.5 | OpenAI | Feb 27, 2025 | $366M |
| 03 | Grok 3 | xAI | Feb 17, 2025 | $218M |
| 04 | Llama 3.1-405B | Meta AI | Jul 23, 2024 | $53M |
| 05 | Llama 4 Behemoth (preview) | Meta AI | Apr 5, 2025 | $45M |
| 06 | GPT-4 (Mar 2023) | OpenAI | Mar 15, 2023 | $37M |
| 07 | Grok-2 | xAI | Aug 13, 2024 | $32M |
| 08 | Gemini 1.0 Ultra | Google DeepMind | Dec 6, 2023 | $31M |
| 09 | Claude 3.5 Sonnet | Anthropic | Jun 20, 2024 | $26M |
| 10 | Nemotron-4 340B | NVIDIA | Jun 14, 2024 | $21M |
| 11 | Mistral Large | Mistral AI | Feb 26, 2024 | $14M |
| 12 | Inflection-2 | Inflection AI | Nov 22, 2023 | $13M |
| 13 | Inflection-2.5 | Inflection AI | Mar 7, 2024 | $12M |
| 14 | Falcon-180B | Technology Innovation Institute | Sep 6, 2023 | $11M |
| 15 | A.X K1 | SK Telecom | Dec 30, 2025 | $10M |
| 16 | Amazon Titan | Amazon | Sep 28, 2023 | $8M |
| 17 | DeepSeek-R1 (May 2025) | DeepSeek | May 28, 2025 | $7M |
| 18 | DeepSeek-R1 | DeepSeek | Jan 20, 2025 | $7M |
| 19 | DeepSeek-V3 | DeepSeek | Dec 24, 2024 | $5M |
| 20 | DeepSeek-V3 (Mar 2025) | DeepSeek | Mar 24, 2025 | $5M |
| 21 | SenseChat | SenseTime | Apr 10, 2023 | $5M |
| 22 | PaLM 2 | May 10, 2023 | $5M | |
| 23 | Claude 2 | Anthropic | Jul 11, 2023 | $5M |
| 24 | GPT-3.5 (davinci-002) | OpenAI | Mar 15, 2022 | $5M |
| 25 | OpenAI Five | OpenAI | Dec 13, 2019 | $4M |
| 26 | Megatron-Turing NLG 530B | Microsoft, NVIDIA | Oct 11, 2021 | $4M |
| 27 | PaLM (540B) | Google Research | Apr 4, 2022 | $3M |
| 28 | Wu Dao 2.0 | Beijing Academy of Artificial Intelligence / BAAI | May 31, 2021 | $3M |
| 29 | EXAONE 1.0 | LG | Dec 14, 2021 | $3M |
| 30 | MegaScale (Production) | ByteDance, Peking University | Feb 23, 2024 | $3M |
| 31 | GPT-3 175B (davinci) | OpenAI | May 28, 2020 | $2M |
| 32 | Arctic | Snowflake | Apr 24, 2024 | $2M |
| 33 | xTrimoPGLM -100B | Tsinghua University, BioMap Research | Jul 6, 2023 | $2M |
| 34 | InternLM | Shanghai AI Lab, SenseTime | Jul 6, 2023 | $2M |
| 35 | Llama 2-70B | Meta AI | Jul 18, 2023 | $1M |
| 36 | BLOOM-176B | Hugging Face, BigScience | Jul 11, 2022 | $996K |
| 37 | Jurassic-1-Jumbo | AI21 Labs | Aug 11, 2021 | $837K |
| 38 | GLM-130B | Tsinghua University | Aug 4, 2022 | $820K |
| 39 | OPT-175B | Meta AI | May 2, 2022 | $734K |
| 40 | Gopher (280B) | DeepMind | Dec 8, 2021 | $641K |
| 41 | Llama 2-34B | Meta AI | Jul 18, 2023 | $600K |
| 42 | Yuan 1.0 | Inspur | Oct 12, 2021 | $592K |
| 43 | Galactica | Meta AI | Nov 16, 2022 | $591K |
| 44 | HyperCLOVA 82B | NAVER, Search Solutions | Sep 10, 2021 | $586K |
| 45 | LLaMA-65B | Meta AI | Feb 24, 2023 | $578K |
| 46 | GLaM | Dec 13, 2021 | $541K | |
| 47 | AlphaGo Master | DeepMind | Oct 19, 2017 | $471K |
| 48 | HyperCLOVA 204B | NAVER | Sep 10, 2021 | $442K |
| 49 | Parti | Google Research | Jun 22, 2022 | $427K |
| 50 | BloombergGPT | Bloomberg, Johns Hopkins University | Mar 30, 2023 | $370K |
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.
Browse all ai models →Other rankings of ai models