DeepSeek is a Chinese AI company that develops language models. In AI Benchmark we track 3 models on DeepSeek with prices, context, speed and usage rates. The highest Intelligence Index belongs to DeepSeek V4.1 Flash 1017 (32,6). The data is updated every day.
Prices are in dollars per 1M tokens: from $0,3 for input at DeepSeek V4.1 Flash 1017 up to the more powerful models. The largest context window is 1 million tokens (DeepSeek-V4.1-Flash), that is, how much text the model holds in a single conversation. Fastest is DeepSeek V4.1 Flash 1017 with about 165 tokens per second. Below are all the reasoning levels (effort) of each model as separate rows: Intelligence Index score, cost per task, input and output price, cache, time to first token and speed. The Intelligence Index is an overall score from independent Artificial Analysis tests. A higher reasoning level usually gives a better result, but uses more tokens and costs more.
Tracked models3with a separate page
Highest Intelligence Index32,6DeepSeek V4.1 Flash 1017
Lowest input price$0,3per 1M tokens, DeepSeek V4.1 Flash 1017
Largest context1 milliontokens, DeepSeek-V4.1-Flash
Fastest model165tokens/sec., DeepSeek V4.1 Flash 1017
Which model from DeepSeek to choose
The choice depends directly on the numbers above. For your task, test with real examples.
All tracked models from DeepSeek
Intelligence Index32,6
Input / 1M$0,3
Output / 1M$1,2
Contextn/a
Speed165 t/s
10,000 requests*$9
DeepSeek V4.1 Flash 1017 is a DeepSeek model published in Artificial Analysis data dated October 2026. Its top level (Max) has an Intelligence Index of 32.6 and ranks 25th out of 82 tracked models. The API price is $0.3 per 1M input and $1,2 for 1…
Intelligence Index23,1
Input / 1M$1,32
Output / 1M$3,96
Contextn/a
Speed68 t/s
10,000 requests*$33
DeepSeek V4 Pro 0813 is a DeepSeek model, listed in Artificial Analysis data dated August 2026. Its top level (Max) has an Intelligence Index of 23.1 and ranks 45th out of 82 tracked models. The API price is $1.32 per 1 million input and $3,96 for 1 M…
Intelligence Indexn/a
Input / 1M$0,3
Output / 1M$1,2
Context1 million
Speedn/a
10,000 requests*$9
DeepSeek-V4.1-Flash is DeepSeek’s current model in the API (name deepseek-flash). It was introduced on 10 September 2026 as an MoE model with 552 billion parameters, 8 billion active at prefill and 16 billion at decode. The weights are under the MIT license.
* Example API spend for 10,000 requests of 1000 input and 500 output tokens each. The actual bill depends on the length of the texts.
Compare models with each other
Compare DeepSeek models side by side
Frequently asked questions about DeepSeek
Which models does DeepSeek tracks AI Benchmark?
Currently we track 3 models: DeepSeek V4.1 Flash 1017, DeepSeek V4 Pro 0813, DeepSeek-V4.1-Flash. Each has its own page with price, context and usage rates. The list and data are updated every day.
Which is the best model for DeepSeek?
Among the tracked models, the highest Artificial Analysis Intelligence Index belongs to DeepSeek V4.1 Flash 1017: 32,6 points. The index is an overall score from several independent tests and does not say which model is best for your specific task.
How much it costs DeepSeek via API?
Prices are per 1M tokens. The cheapest model is DeepSeek V4.1 Flash 1017: $0,3 for input and $1,2 for output. The highest output price is $3,96. Monthly app subscriptions are not included in these numbers.
Which model from DeepSeek is the fastest?
DeepSeek V4.1 Flash 1017 responds fastest among those tracked: about 165 tokens per second according to Artificial Analysis data.
How large is the context of DeepSeek?
The largest context window is 1 million tokens at DeepSeek-V4.1-Flash. This is how much text the model can hold in one conversation.
Are open-weight models the models from DeepSeek?
Open weights among the tracked ones include: DeepSeek-V4.1-Flash.
Sources: Artificial Analysis, the official pages of DeepSeek. Last check: 22.11.2026. To the full AI Benchmark