DeepSeek V4.1 Flash
DeepSeek's current Flash model, called with the name deepseek-flash. It has a 1M-token context and up to 384K tokens of output, reads images, and is the cheaper of DeepSeek's two models.
| Model name | deepseek-flash |
|---|---|
| Served by | DeepSeek-V4.1-Flash-1017 |
| Context | 1M |
| Max output | 384K |
| Image input | Yes |
| Concurrent requests | 2,500 |
| Model | Input, cache hit | Input, cache miss | Output | |
|---|---|---|---|---|
| DeepSeek V4.1 Flash | DeepSeek, peak hours | $0.006 | $0.3 | $1.2 |
| DeepSeek, off-peak | $0.003 | $0.15 | $0.6 | |
| HaryAI | $0.003 | $0.135 | $0.54 |
Get started
Ask an AI to do it
Send this to Codex or Claude Code (or whatever AI assistant you use). Put your key in place of <your HaryAI key> first.
Set up Codex on this computer to use HaryAI.
1. Back up ~/.codex/config.toml (copy it to ~/.codex/config.toml.bak). If it doesn't exist, create it.
2. Add these settings. If a setting with the same name is already there, change its value. Leave everything else as it is.
```toml
model = "deepseek-flash"
model_provider = "haryai"
forced_login_method = "api"
model_reasoning_effort = "high"
[model_providers.haryai]
name = "HaryAI"
base_url = "https://api.haryai.com/v1"
wire_api = "responses"
experimental_bearer_token = "<your HaryAI key>"
```
3. Then tell me to quit Codex completely, reopen it, and send a short message to check that it answers.
4. Also tell me how to undo this: put the backup back and reopen Codex.Set up Claude Code on this computer to use HaryAI.
1. Back up ~/.claude/settings.json (copy it to ~/.claude/settings.json.bak). If it doesn't exist, create it containing {}.
2. Add the entries below to the "env" object in that file. If an entry with the same name is already there, change its value. Keep every other setting, and keep the file valid JSON.
```json
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.haryai.com",
"ANTHROPIC_AUTH_TOKEN": "<your HaryAI key>",
"ANTHROPIC_MODEL": "deepseek-flash[1m]",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek-flash[1m]",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek-flash[1m]",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-flash",
"CLAUDE_CODE_SUBAGENT_MODEL": "deepseek-flash",
"CLAUDE_CODE_EFFORT_LEVEL": "max",
"CLAUDE_CODE_AUTO_COMPACT_WINDOW": "786432",
"CLAUDE_CODE_DISABLE_ARTIFACT": "1"
}
}
```
3. Then tell me to start a new Claude Code session and send a short message to check that it answers.
4. Also tell me how to undo this: put the backup back and start a new session.One click with CC Switch
CC Switch is a free app that switches your tools between providers. Install it from its GitHub releases page. Account page
Write code
- OpenAI format
- https://api.haryai.com/v1
- Anthropic format
- https://api.haryai.com
- Model
- deepseek-flash
Questions
What model name do I use?
deepseek-flash. The old names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work and are served by this model.
Does it think before answering?
It supports both modes. Thinking is on by default and can be switched off.
Can I use it in Codex or Claude Code?
Yes. It's the model DeepSeek's own Codex and Claude Code setup uses by default.
Sources
DeepSeek's figures are from its own docs, checked on 2026-11-15.
DeepSeek at 55% off the official price
Cheaper than DeepSeek's own off-peak price, around the clock.