Market note6 min read
How DeepSeek went from price war to price rise
DeepSeek's cheapest model costs $1.20 per million output tokens at peak, up from $0.28 in April. Every seller's price today, and every release's price and score.
DeepSeek's cheapest model was back at its 2024 price in April.
It now costs three times that.
The full price of DeepSeek's cheapest chat model on its own API, at each change.
- USD per million tokens, blended 3 in to 1 out, log scale
Where these figures come from. The sellers' prices in the first table are the record: each is a list price Vitna read from the seller's own page, confirmed . Everything DeepSeek charged on its own API is quoted from its announcements and price pages, footnoted to the source list at the end. Where a number is our arithmetic, a blend or a ratio, we say so where it appears.
In May 2024 DeepSeek priced V2 at one yuan per million input tokens and set off a price war among China's largest labs.23 Over the next two years it raised and cut its list as the models changed, and by April 2026 its cheapest model cost exactly what V2 had.11 In August 2026 it raised its prices at every hour of the day and began charging double at peak.14
This is what DeepSeek costs now, from DeepSeek and from every seller of its open weights, how the price moved release by release, and what the money bought on Epoch AI's index. For the company and the people behind the models, read how a stock trader came to own DeepSeek.
DeepSeek's prices from its own pages111; the board's blend confirmed ; Epoch AI's index read 20 Nov 2026.
What DeepSeek costs today
The record prices 14 DeepSeek models across 8 sellers, each read from the seller's own pricing page. The cheapest blended 3 in to 1 out million tokens is $0.09, DeepSeek V4 Flash 0731 on DeepInfra. The same weights are not the same price everywhere: DeepSeek V4 Flash 0731 costs 7.3 times as much at Novita AI as at DeepInfra.
| Model, cheapest seller | Input | Output | Blended | Sellers |
|---|---|---|---|---|
| DeepSeek V4.1 Flash 1017Baseten, confirmed 22 Nov 2026 | $0.30 | $1.20 | $0.525 | 4 |
| DeepSeek V4.1 FlashBaseten, confirmed 22 Nov 2026 | $0.30 | $1.20 | $0.525 | 4 |
| DeepSeek V4.1 Flash FastBaseten, confirmed 22 Nov 2026 | $0.60 | $2.40 | $1.05 | 1 |
| DeepSeek V4 Pro 0813Baseten, confirmed 22 Nov 2026 | $1.32 | $3.96 | $1.98 | 3 |
| DeepSeek V4 Flash 0731DeepInfra, confirmed 22 Nov 2026 | $0.06 | $0.18 | $0.09 | 3 |
| DeepSeek V4 FlashDeepInfra, confirmed 22 Nov 2026 | $0.09 | $0.18 | $0.113 | 2 |
| DeepSeek V4 ProDeepInfra, confirmed 22 Nov 2026 | $1.30 | $2.60 | $1.625 | 3 |
| DeepSeek V3.2DeepInfra, confirmed 22 Nov 2026 | $0.26 | $0.38 | $0.29 | 3 |
| DeepSeek V3.1DeepInfra, confirmed 22 Nov 2026 | $0.25 | $0.95 | $0.425 | 3 |
| DeepSeek R1 0528DeepInfra, confirmed 22 Nov 2026 | $0.50 | $2.15 | $0.913 | 3 |
| DeepSeek V3DeepInfra, confirmed 22 Nov 2026 | $0.32 | $0.89 | $0.463 | 1 |
| DeepSeek R1AWS, confirmed 22 Nov 2026 | $1.35 | $5.40 | $2.363 | 1 |
| DeepSeek R1 (Turbo)Novita AI, confirmed 22 Nov 2026 | $0.70 | $2.50 | $1.15 | 1 |
| DeepSeek-OCR 2Novita AI, confirmed 22 Nov 2026 | $0.03 | $0.03 | $0.03 | 1 |
USD per million tokens, each a seller's list price when the collector last read its page, not what anybody paid; blended blended 3 in to 1 out is our arithmetic. A model links to its listing and a seller to its record. Every seller of every model, cached rates included, is on DeepSeek's catalog page.
DeepSeek's own API is not one of the sources the collector reads. Its list, as its pricing page printed it on 18 Oct 2026,1 charges half at night and on weekends:
| Model | Cache hit | Cache miss | Output |
|---|---|---|---|
| deepseek-flash (V4.1 Flash) | $0.003 / $0.006 | $0.15 / $0.30 | $0.60 / $1.20 |
| deepseek-v4-pro (V4 Pro 0813) | $0.022 / $0.044 | $0.66 / $1.32 | $1.98 / $3.96 |
USD per million tokens, off-peak / peak. Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday; every other hour is off-peak.
Every price change, V2 to V4.1
DeepSeek has priced every release on its own API from the day it shipped, and August 2026 was not its first rise. V3 nearly quadrupled output once its launch promotion ended, and V3.1's single list cost more than V3's for anyone not using reasoning. August was the first rise since, and at peak the largest: output on the cheapest model went from $0.28 in July to $1.32, or $0.66 off-peak.14
| Date | Change | Price, per million |
|---|---|---|
| 6 May 2024 | V2 starts a price war. A 236B mixture-of-experts model at one yuan per million input tokens; within three weeks Alibaba, Baidu and others had cut their prices or made models free. 23 | $0.14 in · $0.28 out¥1 and ¥2 on the Chinese list |
| 2 Aug 2024 | Repeated input gets 90% cheaper. Context caching on disk, which DeepSeek said its compressed attention cache made affordable. 4 | $0.014 cache hitagainst $0.14 for a miss |
| 26 Dec 2024 | V3 raises the price. The new model held V2's price for a 45-day promotion, then nearly doubled input and nearly quadrupled output. 5 | $0.27 in · $1.10 outfrom 8 Feb 2025 |
| 20 Jan 2025 | R1 at a 27th of o1. A reasoning model under the MIT licence, priced against OpenAI's o1 at $15 in and $60 out. 67 | $0.55 in · $2.19 outreasoning tokens billed as output |
| 26 Feb 2025 | Night prices. Half off chat and 75% off reasoning from 16:30 to 00:30 UTC, the overnight hours in Beijing. 8 | $0.135 in · $0.55 outboth models, off-peak |
| 21 Aug 2025 | One price for both modes. V3.1 thinks or answers directly from one model and one list; the night discount ended with it. 9 | $0.56 in · $1.68 outfrom 5 Sep 2025 |
| 29 Sep 2025 | Sparse attention, half the price. V3.2-Exp introduced DeepSeek Sparse Attention and cut the list by more than half the same day. 10 | $0.28 in · $0.42 outeffective immediately |
| 24 Apr 2026 | V4 splits into Flash and Pro. A million tokens of context became the default, and the list became two tiers. 11 | Flash $0.14 · $0.28Pro $1.74 in · $3.48 out |
| 22 May 2026 | Pro at a quarter. A launch discount of 75% on Pro became its permanent price when the promotion ended on 31 May. 12 | $0.435 in · $0.87 outV4 Pro |
| 16 Aug 2026 | Peak hours, and a rise. Seven hours a day became peak, at twice the off-peak rate, and both tiers went up at every hour. 1314 | Flash $0.44 · $1.32peak; off-peak half that; Pro $1.32 · $3.96 |
| 10 Sep 2026 | V4.1 Flash takes some of it back. A new architecture with a quarter of the attention cache, and a Flash price below August's. 15 | $0.30 in · $1.20 outpeak; off-peak $0.15 · $0.60 |
What the money bought
Epoch AI scores every DeepSeek release on its Capabilities Index.16 On that index DeepSeek has never held the top. It came closest with DeepSeek-R1 in Jan 2025, 2.5 points under o1. Its latest, DeepSeek V4.1 Flash 1017, sits 9.8 under GPT-6 Astra.
DeepSeek has never held the top of the index.
Each DeepSeek release on Epoch AI's Capabilities Index, against the best model on the index that day.
- DeepSeek's release
- The top of the index that day
| Release | Epoch index | Under the top | The top that day |
|---|---|---|---|
| DeepSeek-V27 May 2024 | 120.8 | 5.0 | GPT-4 Turbo |
| DeepSeek-V326 Dec 2024 | 131.7 | 9.5 | o1 |
| DeepSeek-R120 Jan 2025 | 138.6 | 2.5 | o1 |
| DeepSeek-V3.121 Aug 2025 | 140.9 | 9.1 | GPT-5 |
| DeepSeek-V3.2-Exp29 Sep 2025 | 144.4 | 5.6 | GPT-5 |
| DeepSeek-V4-Pro24 Apr 2026 | 150.2 | 12.6 | GPT-5.5 Pro |
| DeepSeek V4 Pro 081313 Aug 2026 | 156.0 | 8.8 | Claude Opus 5 |
| DeepSeek V4.1 Flash9 Sep 2026 | 157.0 | 12.2 | GPT-6 Astra |
Each release as Epoch AI dates and names it, beside the best model on the index released on or before that day.
One thing the price list cannot show: DeepSeek publishes its weights, so other sellers serve them at their own prices. On the board today DeepSeek V4 Flash 0731 on DeepInfra costs $0.09 blended, under the $0.262 DeepSeek charges off-peak for its cheapest model, so the cheapest way to run a DeepSeek model is not always DeepSeek's own API.
Sources
- 1Price listModels and pricingDeepSeekread
- 2ArchiveAPI pricing, archived 8 May 2024 (both currencies)DeepSeekread
- 3PressChinese tech giants slash prices of language models used to power AI chatbotsReuters, via BusinessDayread
- 4ReleaseDeepSeek API introduces Context Caching on DiskDeepSeekread
- 5ReleaseIntroducing DeepSeek-V3DeepSeekread
- 6ReleaseDeepSeek-R1 ReleaseDeepSeekread
- 7ArchiveAPI pricing, archived 21 January 2025OpenAIread
- 8ArchiveModels and pricing, archived 26 February 2025DeepSeekread
- 9ReleaseDeepSeek-V3.1 ReleaseDeepSeekread
- 10ReleaseIntroducing DeepSeek-V3.2-ExpDeepSeekread
- 11ReleaseDeepSeek V4 Preview ReleaseDeepSeekread
- 12ArchiveModels and pricing, archived 22 May 2026DeepSeekread
- 13ReleaseDeepSeek-V4-Pro GA ReleaseDeepSeekread
- 14ArchiveModels and pricing, archived 16 August 2026DeepSeekread
- 15ReleaseDeepSeek-V4.1-Flash: Smarter, Faster, More EfficientDeepSeekread
- 16DatasetEpoch Capabilities Index (CC BY 4.0)Epoch AIread
More posts
Why a Google SERP API now returns five results, not ten
Google now unlocks each result link a little later than the one above it, the tenth about 24 seconds after the page. Five results arrive before the long waits: 2,273 of 2,500 live searches answered, 12 at a time, at a median of 4.19 seconds.
Claude API pricing: how Opus fell from $75 to $20
A million tokens of Opus output cost $75 in May 2025. Opus 5.5 launched at $20 in September. Every Claude model's price today, every change since, and what one coding session costs.
Who owns the AI companies, and who still holds the votes
Elon Musk owns under half of SpaceX, home of xAI, and casts 82% of its votes. Microsoft owns about a quarter of OpenAI and names none of its directors. How the money and the votes came apart at 13 AI labs.
How Beam 1.0 got past its tokens-per-second ceiling
Beam 1.0 runs at 348.8 tokens a second across its hosts and 537.3 on the fastest. One token a pass stops near 300. What sets that ceiling, and the ways past it.
How a stock trader came to own DeepSeek
Liang Wenfeng still controls DeepSeek. His hedge fund paid for it for three years, and the outside investors who bought in this June hold under 9%.
