DeepSeek-V4.1-Flash-1017: What Changed and How to Call It via API

On 17 October 2026 DeepSeek released DeepSeek-V4.1-Flash-1017, a post-training update of DeepSeek-V4.1-Flash, which first shipped on 10 September. DeepSeek serves it under the same API name as before, deepseek-flash, so existing apps picked it up without a code change.

This post covers what changed for API users, which model ID to send on cheapaimodel, how to check which version answered a request, and what it costs. Prices were checked on 2 November 2026.

What changed

According to DeepSeek, the 1017 update scores better on its benchmarks, finishes tasks faster and writes shorter outputs than the September release. DeepSeek also says it is optimized for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy. Those are DeepSeek’s own claims; run your own evaluation on your prompts before you rely on them.

For API users, nothing else moved:

• Price: unchanged. DeepSeek still lists $0.30 per million input tokens and $1.20 per million output tokens at peak, and half that off-peak.
• Limits: 1M-token context window and up to 384K output tokens.
• Inputs: text and images in, text out.
• Thinking: runs in thinking mode by default, with a non-thinking mode available.

Which model ID to send

On cheapaimodel, three model IDs currently reach the 1017 release. They are billed the same way.

Model ID	Use it when
deepseek-v4-1-flash	You want the latest DeepSeek-V4.1-Flash. Recommended.
deepseek-v4-1-flash-1017	Your config or logs name the release explicitly.
deepseek-v4.1-flash-1017	You copied the name from the model field of a response.

deepseek-v4-1-flash follows DeepSeek: when DeepSeek ships the next update, this name moves to it, the same way deepseek-flash does on DeepSeek’s own API. The two -1017 names exist so that a name you read in a response also works in a request. We can’t promise they will keep pointing at 1017 after DeepSeek’s next update, so don’t treat them as a pinned version.

If your API key has an allowed-models list, it only accepts the names written in it. Add the -1017 names to the list if you plan to send them.

Call it

The endpoint is OpenAI-compatible. Point base_url at https://api.cheapaimodel.ai/v1 and send your cheapaimodel key.

curl https://api.cheapaimodel.ai/v1/chat/completions \
  -H "Authorization: Bearer $CHEAPAIMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-1-flash",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

With the OpenAI Python SDK:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.cheapaimodel.ai/v1",
    api_key=os.environ["CHEAPAIMODEL_API_KEY"],
)

response = client.chat.completions.create(
    model="deepseek-v4-1-flash",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(response.model)  # deepseek-v4.1-flash-1017
print(response.choices[0].message.content)

The Quickstart covers keys and other SDKs.

Check which version answered

The model field of every response names the release that handled the request. It is different from the name you sent, and that is expected:

{
  "object": "chat.completion",
  "model": "deepseek-v4.1-flash-1017",
  "choices": [ … ]
}

The same name appears in every chunk of a streaming response and in the Responses API (/v1/responses). Keep sending deepseek-v4-1-flash; you don’t need to switch your requests to the name in the response.

One side effect: tools that look up prices by the response’s model field, LiteLLM among them, won’t find deepseek-v4.1-flash-1017 in their price tables. Map it to deepseek-v4-1-flash or price it by the model you requested.

The version section of the model page lists the current release, the name in responses and the accepted aliases, and is updated when DeepSeek ships a new release.

What it costs

DeepSeek prices this model by time of day. Peak hours are 09:00–12:00 and 14:00–18:00 Beijing time on weekdays (01:00–04:00 and 06:00–10:00 UTC). Every other hour, including weekends, is off-peak.

Per 1M tokens	Input	Cache read	Output
DeepSeek list price, peak	$0.30	$0.006	$1.20
DeepSeek list price, off-peak	$0.15	$0.003	$0.60
cheapaimodel, peak	$0.210015	$0.004200	$0.840061
cheapaimodel, off-peak	$0.105008	$0.002100	$0.420031

A workload of 1M input tokens and 200K output tokens costs:

• $0.54 at DeepSeek’s peak price, or $0.27 off-peak.
• $0.378027 on cheapaimodel at peak, or $0.189014 off-peak.

Off-peak on cheapaimodel is 65% below DeepSeek’s peak list price, the same figure the model page shows. Billing is prepaid in US dollars, with no subscription. DeepSeek API pricing has more worked examples.

Turn thinking off when you don’t need it

DeepSeek-V4.1-Flash-1017 thinks by default, and reasoning tokens are billed as output. For translation, extraction or short replies, send "reasoning_effort": "none" to switch thinking off; in our tests it brought reasoning tokens to zero and cut output tokens by about 80% on a translation job. How to set DeepSeek V4.1 Flash’s reasoning effort shows the request and the trade-offs.

FAQ

Do I need to change my code for 1017? No. If you already send deepseek-v4-1-flash, your requests have been going to 1017 since the switch. Only the model field in responses changed.

Can I keep using the September release? Not through cheapaimodel. We serve the release DeepSeek currently serves under deepseek-flash, and that is 1017.

Is 1017 more expensive? No. DeepSeek kept the same list prices, and our prices didn’t change with the release.

Why does the response say a different model name? The response reports the exact release that answered, while the name you send follows the latest release. Both refer to the same model.

Ready to try it? Get an API key or see the DeepSeek-V4.1-Flash model page.