Latest
  1. Johns Hopkins Bayview tells 3,182 patients an employee put their health data into a public AI tool
  2. China’s AI compute access expected to double in 2027, with US-designed chips about 65 per cent
  3. UK commits £100m to national AI Biosecurity Lab
  4. Moonshot AI releases Kimi K3.1 Pro with 1-million-token context window
  5. Residents in Newport and Cardiff oppose hyperscale datacentres near homes
  6. Study: Grok 4.7 picked Republicans in 92% of forced midterm answers, GPT-6.1 Sol picked Democrats in 97%
  7. Attackers exploiting the PHP flaws hand the break-in to AI agents, Unit 42 says
  8. Elvis Presley’s rightsholders sue Anthropic over training on 274 songs

Latest / Frontier models and safety

Model releaseFrontierComputeChinaConfirmed

DeepSeek releases V3, a 671B open-weights model trained on 2.788M H800 GPU hours

See how this story connects

Source: DeepSeek API DocsAlso reported by DeepSeek-V3 Technical Report (arXiv)

Chinese lab DeepSeek released DeepSeek-V3 on 26 December 2024, a mixture-of-experts model with 671 billion parameters, 37 billion active per token, and published the weights and a technical report. The report says pre-training used 14.8 trillion tokens, full training took 2.788 million H800 GPU hours, and the model outperforms other open-source models and is comparable to leading closed ones. Standard API prices from 8 February 2025 were 2 yuan per million input tokens on a cache miss and 8 yuan per million output tokens.

Why it matters

Developers anywhere could download a Chinese model its makers said rivalled leading closed systems, with a disclosed training budget of 2.788 million H800 GPU hours.

When
26 Dec 2024
ID
WTA-N-2024-0050
Where
China
Who
DeepSeek, NVIDIA

Summary by wheresthe.ai, from DeepSeek API Docs and the reporting credited above. Checked against the source on 9 Oct 2026.

Read the original at DeepSeek API Docs

Source copy: download the page as saved on 18 Oct 2026 (PDF). Also held by the Internet Archive.

Next stop on this Storyline: Led toDeepSeek releases R1 reasoning model with open weights under MIT licence

Cite this story

WTA-N-2024-0050: “DeepSeek releases V3, a 671B open-weights model trained on 2.788M H800 GPU hours.” wheresthe.ai, 26 Dec 2024. https://wheresthe.ai/d/cn-deepseek-v3-release-2024-12/ (source: DeepSeek API Docs).

Spotted an error? Email [email protected].