Compare models

DeepSeek V3 vs GPT-4o

Every figure below carries the source that reported it and the moment it was observed. Values that failed validation are withheld rather than shown, and a dash means no source we track carries that field for that model.
DeepSeek V3 vs GPT-4o, compared across 14 tracked fields. Each value shows the source that reported it and when.
FieldDeepSeek V3GPT-4o
Downloads, 30 days
1,163,331Synced from sourceCurrent1h ago
—
Likes
4,308Synced from sourceCurrent1h ago
—
Context window
163,840 (highest of the compared values)Synced from sourceCurrent1h ago
128,000Synced from sourceCurrent1h ago
Input, $/M tokens
0.26 (lowest of the compared values)Synced from sourceCurrent1h ago
2.5Synced from sourceCurrent1h ago
Modality
textSeeded, unreviewedwritten 1mo ago
textSeeded, unreviewedwritten 1mo ago
Licence
deepseekSeeded, unreviewedwritten 1mo ago
proprietarySeeded, unreviewedwritten 1mo ago
Gated
noSynced from sourceCurrent1h ago
—
Last updated
2025-03-27Synced from sourceCurrent1h ago
—
Library
transformersSynced from sourceCurrent1h ago
—
Task
text-generationSynced from sourceCurrent1h ago
—
Hugging Face
deepseek-ai/DeepSeek-V3Synced from sourceCurrent1h ago
—
OpenRouter id
deepseek/deepseek-chatSynced from sourceCurrent1h ago
openai/gpt-4oSynced from sourceCurrent1h ago
Output, $/M tokens
1.03Synced from sourceCurrent1h ago
10Synced from sourceCurrent1h ago
Vendor
DeepSeekSeeded, unreviewedwritten 1mo ago
OpenAISeeded, unreviewedwritten 1mo ago

Sources: Hugging Face Hub, OpenRouter model catalogue. Most recent observation 1h ago. Where a row is marked, ▲ is the highest of the values shown and ▼ the lowest — arithmetic on the figures above, not a ranking or a recommendation. Rows where neither extreme is meaningful are left unmarked rather than given a direction they do not have.

Change the comparison

Remove one, or open the directory to pick a different set. Up to 3 at a time.


Every figure on this site resolves to the source that reported it, the moment it was observed, and how it was measured. A value that fails validation is withheld rather than shown — withheld means we refused to publish it, not that it is zero.