deepseek-flash vs hy4
- deepseek-flash
- 1 advantage
- hy4
- No edge in the figures
How to choose
deepseek-flashReads images
hy4No edge in the figures
deepseek-flash
By DeepSeek
hy4
By TencentNew in the last two weeks
Where they differ
deepseek-flashhy4
On this site
100%83 testsAvailability, this week's tests—
$1.18Reference price per million output tokens$1.25
#11Place in this week's rankingUnranked
$0.29Reference price per million input tokens$0.42
The model itself
384K tokensLongest reply64K tokens
YesReads imagesNo
May 2025Knowledge cutoffNot published
14 Sep 2026Released4 Nov 2026
$0.0059Reference price per million cache-read tokens$0.021
Where they match
- Context length1M tokens
- Calls toolsYes
- Thinks before answeringYes
- Conversations used for trainingUsed to train and improve the provider’s models
- Open weightsYes
Availability over 3 days counts every call made through the site; this week's availability counts every call over the last 7 days, scheduled tests included, and response time comes from the site's own scheduled tests. Models marked as free are counted in calls; paid models are charged by usage, in US dollars.
Compare next
deepseek-flash against other models
- vs gpt-5.6-luna$1.20Reference price per million output tokens
- vs mimo-v2.6-pro$0.87Reference price per million output tokens
- vs mistral-small-2603$0.60Reference price per million output tokens
- vs gpt-6-luna$0.50Reference price per million output tokens
- vs claude-haiku-5.5$0.50Reference price per million output tokens
- vs deepseek-v4.1-flash$1.20Reference price per million output tokens
hy4 against other models
- vs deepseek-v4.1-flash$1.20Reference price per million output tokens
- vs gpt-5.6-luna$1.20Reference price per million output tokens
- vs mimo-v2.6-pro$0.87Reference price per million output tokens
- vs mistral-small-2603$0.60Reference price per million output tokens
- vs gpt-6-luna$0.50Reference price per million output tokens
- vs hy4-flash$0.28Reference price per million output tokens
Figures come from real calls made through this site. Models are served by third-party providers and may slow down or be withdrawn; this page follows those changes.