DeepSeek V4.1 Flash 1017 Is Now on Risphai: What's New, Benchmarks and Pricing
DeepSeek V4.1 Flash 1017 is live on Risphai and now codes on the Balanced and Quality presets. See what changed, the benchmarks, reasoning levels, pricing and how to try it free.
By Risphai
Benchmark scores reported by DeepSeek for DeepSeek V4.1 Flash 1017 compared with the previous DeepSeek V4.1 Flash.
DeepSeek V4.1 Flash on Risphai now runs on DeepSeek V4.1 Flash 1017, the update DeepSeek released on October 17, 2026. DeepSeek reports better scores on all 11 benchmarks it published, with faster task completion and shorter outputs. It is now also the default coding model on the Balanced and Quality presets, which cuts our estimate for an average Balanced coding request from about 30 to about 15 credits. The name and price per request stay the same, and you can try it on the Free plan.
DeepSeek calls 1017 a minor update to DeepSeek V4.1 Flash, but the benchmark gains are large for a point release, especially in agentic coding. Because of those gains and its low price, we made it the default for coding on every preset except Extreme Quality.
Stronger coding and agents. The biggest gains are on agentic coding and automation benchmarks, such as Terminal-Bench 4.0, ProgramBench and Apex Bench.
Better reasoning and vision. Scores also rise on Humanity's Last Exam and on the visual benchmarks Chartography, BabyVision and ZeroBench.
Shorter outputs and faster tasks. DeepSeek says the model finishes tasks in less time and with shorter outputs, which can lower the cost per task.
Whole projects from one request. DeepSeek trained the update with reinforcement learning and self-feedback to build complete projects, and showed demos such as a web racing game and 3D scenes.
Same model family. It keeps the 1M-token context window, image input, tool calling and adjustable reasoning of DeepSeek V4.1 Flash.
DeepSeek V4.1 Flash 1017 Benchmarks
These are the scores DeepSeek published for the 1017 version and the previous DeepSeek V4.1 Flash. We haven't run these benchmarks ourselves.
Benchmark
V4.1 Flash
V4.1 Flash 1017
Change
HLE (no tools)
36.8
42.6
+5.8
Terminal-Bench 4.0
31.2
42.3
+11.1
DeepSWE v1.1
74.2
77.0
+2.8
ProgramBench
20.3
29.1
+8.8
NL2Repo-Bench
65.4
68.2
+2.8
Automation-Bench v1.0.6
54.8
60.5
+5.7
Agents' Last Exam
31.8
35.4
+3.6
Apex Bench v1.1
64.6
72.1
+7.5
Chartography (with tools)
78.9
83.0
+4.1
BabyVision (with tools)
89.6
92.5
+2.9
ZeroBench-main (with tools)
49.0
51.0
+2.0
DeepSeek used the Claude Code harness for Terminal-Bench 4.0 and the Mini-SWE-V2 harness for DeepSWE v1.1. On the text-only subset of HLE, the score rose from 39.1 to 44.0. DeepSeek's full table also compares other models. Benchmarks measure specific tasks, so results in your own work can be different.
What Changes for You on Risphai
Nothing you need to do. DeepSeek V4.1 Flash keeps its name in the model list, and every DeepSeek V4.1 Flash option now uses the 1017 version:
Your settings stay. Saved model choices and existing conversations keep working.
New coding default. The Balanced preset now codes with DeepSeek V4.1 Flash Thinking (High) and the Quality preset with DeepSeek V4.1 Flash Thinking (Max). Both used GPT-6.1 Sol before, which stays the default on Extreme Quality.
Cheaper coding on Balanced. Our estimate for an average coding request on Balanced drops from about 30 to about 15 credits. The Pricing page shows the new estimate.
Same price. Credits per request didn't change with the update.
Shorter answers can cost less. You pay for the compute and tokens a response uses. If the model writes shorter answers for the same task, as DeepSeek reports, the response can use fewer credits.
Where Risphai Uses DeepSeek V4.1 Flash
Risphai picks a model for each request based on the task and your quality preset. DeepSeek V4.1 Flash is the default here:
Request
Cost Efficient
Fast
Balanced
Quality
Extreme Quality
Reasoning
DeepSeek V4.1 Flash (High)
DeepSeek V4.1 Flash (Low)
Other model
Other model
Other model
Coding
DeepSeek V4.1 Flash (High)
DeepSeek V4.1 Flash (Low)
DeepSeek V4.1 Flash (High)
DeepSeek V4.1 Flash (Max)
GPT-6.1 Sol
The Free plan uses Cost Efficient, so free reasoning and coding requests use DeepSeek V4.1 Flash 1017 by default. The other presets are available on Paid plans.
If the default model is unavailable or your balance doesn't cover it, Risphai automatically uses another model so your request still finishes.
DeepSeek V4.1 Flash Reasoning Levels
DeepSeek V4.1 Flash can answer right away or think first. On Risphai you get four levels:
Level
Where
Best for
No thinking
Chat and function calling
Quick answers and everyday questions
Low
Reasoning and coding
Simple code and fast reasoning
High
Reasoning and coding
Most coding and math tasks
Max
Reasoning and coding
The hardest problems, when time and credits matter less
Higher levels let the model think longer, which usually takes more time and credits. We removed the XHigh level for DeepSeek V4.1 Flash because DeepSeek's documentation treats it the same as High. If you had XHigh selected, Risphai now uses High automatically.
Ask a coding question. On the Cost Efficient, Fast, Balanced and Quality presets, Risphai sends it to DeepSeek V4.1 Flash automatically. Reasoning questions use it on Cost Efficient and Fast.
Or choose it yourself. In Advanced mode on Paid plans, open the model settings in the chat box and select DeepSeek V4.1 Flash for chat or function calling, or DeepSeek V4.1 Flash Thinking with Low, High or Max for reasoning and coding.
Prompts to Try
DeepSeek says the 1017 version is especially good at building complete projects and agentic tasks. These requests are a good start:
"Build a small racing game that runs in the browser in a single HTML file."
"Find the bug in this function and explain the fix step by step." (with your code)
"Write a Python script that renames all photos in a folder by the date they were taken."
"Solve this probability problem and check your answer another way."
"What does this chart show? List the three most important trends." (with a screenshot)
When Risphai writes runnable HTML, CSS and JavaScript, you can run it in the chat in a sandboxed preview and ask for changes.
DeepSeek V4.1 Flash Pricing on Risphai
The price is the same as before the update:
Provider
Price
First provider
0.1 to 1 credits per request on average, depending on the compute used
Fallback provider
Up to 150 credits per million input tokens and 600 credits per million output tokens
Like every response, the total also includes the processing charge described in the Terms. The exact credits are shown for each response, and current prices are on the Information page.
With DeepSeek V4.1 Flash as the coding default on Balanced, we now estimate an average coding request there at about 15 credits instead of about 30. Your actual cost depends on how much code the model writes and how long it thinks.
The Free plan includes 100 credits per month, enough to try DeepSeek V4.1 Flash 1017 on reasoning and coding. Paid plans include 3,000 to 20,000 credits per month, the higher quality presets and model selection in Advanced mode. Compare plans on the Pricing page.
Privacy
DeepSeek V4.1 Flash works in Privacy Mode, which keeps your prompts and responses from being stored by the AI provider. The Privacy Policy explains how your data is processed.
Try DeepSeek V4.1 Flash 1017
DeepSeek V4.1 Flash 1017 is live on Risphai now. Ask a reasoning or coding question on the Free plan, or upgrade on the Pricing page to code with it on Balanced and Quality and to choose the model and its reasoning level yourself. All changes in this release are in the v0.1.4 release notes.
Frequently Asked Questions
What is DeepSeek V4.1 Flash 1017?
DeepSeek V4.1 Flash 1017 is the October 17, 2026 update of DeepSeek V4.1 Flash. DeepSeek describes it as a minor update with better results in reasoning, coding, agentic tasks and visual benchmarks, plus faster task completion and shorter outputs. The open weights are published on Hugging Face as DeepSeek-V4.1-Flash-1017 under the MIT License.
Does Risphai use DeepSeek V4.1 Flash 1017?
Yes. Every DeepSeek V4.1 Flash option on Risphai now runs on the 1017 version. The model keeps its name in the model list, so you don't need to change any settings, and your saved choices and conversations keep working.
Is DeepSeek V4.1 Flash 1017 better than DeepSeek V4.1 Flash?
In the benchmark table DeepSeek published, the 1017 version scores higher than the previous V4.1 Flash on all 11 listed benchmarks, for example 42.3 instead of 31.2 on Terminal-Bench 4.0 and 29.1 instead of 20.3 on ProgramBench. These are DeepSeek's own results, and real-world results depend on the task.
Can I use DeepSeek V4.1 Flash 1017 for free?
Yes. The Free plan includes 100 credits per month and uses the Cost Efficient preset, which sends reasoning and coding requests to DeepSeek V4.1 Flash Thinking (High) by default. Paid plans add 3,000 to 20,000 credits per month and let you choose DeepSeek V4.1 Flash and its reasoning level yourself in Advanced mode.
Which model does Risphai use for coding?
Since October 18, 2026, DeepSeek V4.1 Flash 1017 is the default coding model on the Cost Efficient, Fast, Balanced and Quality presets: Thinking (High) on Cost Efficient and Balanced, Thinking (Low) on Fast and Thinking (Max) on Quality. The Extreme Quality preset uses GPT-6.1 Sol.
How many credits does a coding request use?
On the Balanced preset, we estimate an average coding request at about 15 credits, down from about 30 before DeepSeek V4.1 Flash became the default. The actual cost depends on how much code the model writes and how long it thinks, and the exact credits are shown for each response.
How much does DeepSeek V4.1 Flash 1017 cost on Risphai?
The price did not change with the update. With the first provider, a request costs on average 0.1 to 1 credits, depending on the compute used. If the fallback provider is used, the maximum is 150 credits per million input tokens and 600 credits per million output tokens. Every response also includes the processing charge, and the exact credits are shown for each response.
Which reasoning levels does DeepSeek V4.1 Flash support on Risphai?
Four: no thinking for chat and function calling, and Low, High and Max thinking for reasoning and coding. Higher levels let the model think longer before it answers, which helps with harder problems but usually takes more time and credits.
What happened to the XHigh reasoning level?
We removed XHigh for DeepSeek V4.1 Flash because DeepSeek's API documentation treats xhigh the same as high, so it didn't give you anything High doesn't. If you had XHigh selected, Risphai now uses High automatically.
Can DeepSeek V4.1 Flash 1017 understand images?
Yes. DeepSeek V4.1 Flash accepts text and images, so you can attach screenshots, charts or photos and ask about them.
How large is the DeepSeek V4.1 Flash 1017 context window?
DeepSeek lists a context window of 1 million tokens for DeepSeek V4.1 Flash 1017, the same as the previous version.
Is DeepSeek V4.1 Flash 1017 open source?
The weights are open and published on Hugging Face under the MIT License, according to DeepSeek's model card. On Risphai you don't need to set anything up; just choose the model or let Risphai pick it for you.
