DeepSeek V4.1 Flash is one of the most searched models in our catalog right now, and for good reason: it combines a 1M-token context window, open weights, thinking and non-thinking modes, and a price that is among the lowest for a frontier-class coding model. It is also free on several routes. This guide explains where, what the free routes cost you in data and limits, and what the paid API charges once you outgrow them. Prices and terms below come from DeepSeek's official documentation (page dated November 10, 2026) and from the catalog data on the DeepSeek V4.1 Flash page, last refreshed November 12, 2026.
What DeepSeek V4.1 Flash is
According to DeepSeek's model table, the current Flash model is DeepSeek-V4.1-Flash-1017. In the API you select it with the model name deepseek-flash. The older names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work, but DeepSeek says the models behind them are retired and their requests are served by V4.1 Flash and billed at the Flash price.
| Property | Value |
|---|---|
| Model name in the API | deepseek-flash |
| Context length | 1M tokens |
| Maximum output | 384K tokens |
| Thinking mode | Supports both non-thinking and thinking (default) |
| Tool calls, JSON output, Responses API, Anthropic API format | Supported |
| Vision | Supported on Flash; not supported on V4 Pro |
| FIM completion (beta) | Non-thinking mode only |
| Concurrency limit | 2,500 (V4 Pro: 500) |
| Weights | Open |
On our catalog page the model scores 39.5 on the Artificial Analysis Intelligence Index (reasoning, max effort), with 84.0% on the long-context benchmark and 39.2% on Humanity's Last Exam. Those figures are third-party measurements, which we show as published and never estimate ourselves.
Where it is free
The catalog currently lists three free routes for DeepSeek V4.1 Flash. Each is a separate endpoint with its own limits, so check the terms before you build on one.
| Provider | Model ID | What the provider says |
|---|---|---|
| OrcaRouter | deepseek/deepseek-v4.1-flash-free | Free tier |
| NVIDIA Build | deepseek-ai/deepseek-v4.1-flash | Free serverless inference for development; limits and availability may change. No card required |
| OpenCode Zen | deepseek-v4.1-flash-free | Free for a limited time |
The catalog shows the model as free since September 21, 2026. Free routes come and go, so the page records the date each model first appeared as free and updates every few hours.
OpenCode Zen: free, with a privacy note in your favor
OpenCode's documentation lists DeepSeek V4.1 Flash Free among its limited-time free models and states that its provider follows a zero-retention policy and does not use your data for model training. Not every free model on that list is as strict: some stealth and trial models explicitly collect data to improve themselves, and the NVIDIA free endpoints are marked trial use only, logged, and not for confidential data. Read the privacy section of the provider docs before sending anything private. See our OpenCode Zen catalog for the full list and our experiment with free OpenCode models.
NVIDIA Build: the no-card route
NVIDIA's hosted endpoints are meant for prototyping. They are free without a card, but the terms say limits and model availability can change, and the free service is not a production SLA. For learning, evaluation, and weekend projects it is the easiest way to try DeepSeek V4.1 Flash without payment details. See NVIDIA Build in the catalog.
OrcaRouter and the OpenAI-compatible quick start
OrcaRouter exposes the free route through an OpenAI-compatible endpoint, so any client that lets you change the base URL works: Cursor, Cline, OpenCode, Open WebUI, or a few lines of Python.
from openai import OpenAI
client = OpenAI(
base_url="https://api.orcarouter.ai/v1",
api_key="YOUR_ORCAROUTER_API_KEY",
)
reply = client.chat.completions.create(
model="deepseek/deepseek-v4.1-flash-free",
messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)If you would rather use DeepSeek directly, the base URL is https://api.deepseek.com for the OpenAI format and https://api.deepseek.com/anthropic for the Anthropic format. The Anthropic-format endpoint is what lets tools built for Claude, such as Claude Code, point at DeepSeek models. DeepSeek's docs include integration guides for Claude Code, Codex, OpenCode, and Hermes.
What the paid API costs
Free routes are generous for experimentation, but they are rate-limited and can disappear. DeepSeek's own pricing is low enough that many teams move to it quickly. The table shows prices per 1M tokens for V4.1 Flash and, for comparison, V4 Pro.
| Price per 1M tokens | Flash off-peak | Flash peak | V4 Pro off-peak | V4 Pro peak |
|---|---|---|---|---|
| Input (cache hit) | $0.003 | $0.006 | $0.022 | $0.044 |
| Input (cache miss) | $0.15 | $0.30 | $0.66 | $1.32 |
| Output | $0.60 | $1.20 | $1.98 | $3.96 |
Off-peak rates are exactly half of peak rates. DeepSeek defines peak hours as 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, excluding Chinese public holidays. Everything else, including full weekends and Chinese public holidays, is off-peak. If your jobs are batch work, schedule them for the off-peak window and your bill halves.
A worked example
Suppose a coding agent runs a task that sends 2M input tokens, of which 80% hit the cache, and receives 200K output tokens. At off-peak rates:
- Cache-hit input: 1.6M tokens × $0.003 per 1M = $0.0048
- Cache-miss input: 0.4M tokens × $0.15 per 1M = $0.06
- Output: 0.2M tokens × $0.60 per 1M = $0.12
- Total: about $0.185
The same task at peak rates costs about $0.37. This is an illustration of the arithmetic, not a benchmark of real agent sessions, which vary widely in how much they cache. The practical lesson is that cache hits are almost free, so keep the system prompt and repository context stable between requests.
DeepSeek also sells standalone Search and Fetch APIs at $0.75 per 1,000 successful requests each. Those are separate from model pricing.
DeepSeek inside OpenCode Go
If you use OpenCode, the Go subscription ($10/month; Go Plus $40/month) includes DeepSeek V4.1 Flash and V4 Pro among its open coding models. OpenCode's documentation notes a region setting: by default, paid endpoints that train on request data are blocked and allowed regions are Europe and the United States, and to use the DeepSeek models an owner or admin has to select Global under Regions in Settings, Privacy. In other words, the DeepSeek models are served in a way that is only enabled when you allow the Global region. Read our explainer on OpenCode Zen and Go before you choose.
Choosing a route
| Situation | Best route |
|---|---|
| Trying the model, no card, no commitment | NVIDIA Build |
| Using it inside OpenCode with a strict no-training policy | OpenCode Zen free route |
| Any OpenAI-compatible client, free tier | OrcaRouter |
| Production traffic with predictable limits | DeepSeek's own API, off-peak scheduling where possible |
| Heavy daily agent use at a flat price | OpenCode Go or a pay-as-you-go budget with caching |
Limits and caveats
- Free means rate-limited. Expect per-minute and daily caps. Handle HTTP 429 with backoff and keep a fallback model.
- Free can end. OpenCode describes its free models as free for a limited time. Do not build a product on a free route without a paid fallback.
- Data policy differs by route. Zero-retention on one route does not apply to another.
- Region and compliance. If your data must stay in a specific region, confirm where the route is hosted before you use it.
- Benchmarks are third-party. A leaderboard score does not predict how the model handles your codebase. Test on your own tasks.
FAQ
Is DeepSeek V4.1 Flash really free?
Yes, on at least three routes in our catalog as of November 12, 2026: OrcaRouter, NVIDIA Build, and OpenCode Zen. Each has its own limits and terms. DeepSeek's own API is paid, at the prices above.
What is the model name in the DeepSeek API?
Use deepseek-flash. The legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted and served by V4.1 Flash.
Does it support a 1M-token context?
Yes, DeepSeek lists a 1M context length and a 384K maximum output for the Flash model. Free routes may cap context lower, so check each provider's row on the model page.
When is DeepSeek cheapest?
Off-peak hours, which DeepSeek prices at half the peak rate. Peak is 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, excluding Chinese public holidays.
Can I use it with Claude Code?
DeepSeek provides an Anthropic-format endpoint and a Claude Code integration guide in its docs. Check the guide for the exact environment variables, which can change.
Where do I compare it with other free models?
The catalog lists every free route, and our guide to the best free coding models compares models by benchmark and availability.
