DeepSeek V3 (Dec '24)

By DeepSeek · released 2024-12-26 (21 months) · snapshot 2026-10-24

Off the frontierOn frontier at launch dayopen weights

Intelligence index: 8.50
Cost to run the AA index: $25.4
Price per 1M tokens (in / out): $0.32 / $0.89
End-to-end time: 11.9s

Better than…

Is DeepSeek V3 (Dec '24) on the Pareto frontier, and what beats it?

Off the frontier

No. GPT-6 Luna (Low) is at least as smart for the same money or less.

Next step up: GPT-6 Luna (Medium), smarter but dearer.

DeepSeek V3 (Dec '24) vs GPT-6 Luna (Low)DeepSeek V3 (Dec '24) vs GPT-6 Luna (Medium)

#	Model	Creator	Intelligence	Total cost	E2E total	Released	Frontier
1	DeepSeek V3 (Dec '24)	DeepSeek	8.50	$25.4	11.9s	2024-12-26
2	GPT-6 Luna (Low)	OpenAI	21.50	$10.6	6.60s	2026-09-22	Frontier
3	GPT-6 Luna (Medium)	OpenAI	29.90	$31.2	8.25s	2026-09-22	Frontier

Open in the picker Daily check URL: 200 on the frontier, 409 once beaten

Matchup

Compare head to head

Suggested: what beats it, its effort levels, and frontier models near its price.

Compare DeepSeek V3 (Dec '24) with measured models

Release

What the DeepSeek V3 release did to the frontier on 2024-12-26

1 of 1 effort levels reached the frontier at launch (DeepSeek V3 (Dec '24)). Impact 0, did not move the frontier; 46 releases this past year did.

Release page: launch-day frontier and every effort level

Deciding

Should you use DeepSeek V3 (Dec '24)?

Only when it fits your setup: GPT-6 Luna (Low) is at least as smart for the same money or less, so DeepSeek V3 (Dec '24) is off the value frontier today.

What is better and cheaper than DeepSeek V3 (Dec '24)?

GPT-6 Luna (Low) — at least as smart for the same money or less.

How much does DeepSeek V3 (Dec '24) cost?

$0.32/$0.89 in/out per 1M tokens, and $25.4 to run the full Artificial Analysis Intelligence Index, the honest all-in number for reasoning models.

Is DeepSeek V3 (Dec '24) on the Pareto frontier?

No: GPT-6 Luna (Low) beats it on intelligence, cost, or both.

Frontier membership by axis pair

Other axis pairs where DeepSeek V3 (Dec '24) is Pareto-optimal.

Y axis	X axis	On the frontier
AA intelligence index	AA total cost to run (USD)
AA intelligence index	End-to-end total seconds
AA intelligence index	Output speed tokens/sec

Agents: Accept: text/markdown or ?format=md returns this page as Markdown. JSON: /api/v1/models/deepseek-v3 · Markdown: /api/v1/models/deepseek-v3?format=md · llms.txt · agent skill

Is DeepSeek V3 (Dec '24) on the Pareto frontier, and what beats it?

Off the frontier