Qwen3.7 Plus
Mistral Medium 3.5
763B- input
- $0.15
- cached
- $0.003
- output
- $0.60
- Context
- 1Mtokens
- Listed
- 3w ago
- Size
- 763Bparameters
Cost/1M tokens
DeepSeek-V4.1-Flash is the smallest model in DeepSeek's new architecture family, taking over from the retired V4-Flash and V4-Flash-Vision-Exp, with native visual understanding. The launch headline calls it smarter, faster, and more efficient: DeepSeek claims its KV cache needs just a quarter of the HBM and an eighth of the SSD storage of the previous generation, cutting the cache-hit charges that make up a large share of agent costs, and cites tests putting it ahead of its flagship V4-Pro. It is built for agents and coding, with partners WorkBuddy and OpenCode supporting it.
- Multimodal model
- Open weights · 763B
- Accepts images
- 1M context
- Reasoning
- Tools
- Structured output
How it compares
Score vs. costArena WebDev
Each model at its best score · Dashed: the best score at each cost
- DeepSeek
- Google DeepMind
- OpenAI
- Anthropic
- Z.ai
- Meta
- xAI
- Moonshot AI
- Xiaomi
- Upstage
Level
Mid-tierThe lab's everyday default: strong on most tasks at a mid-range price.
Vs. your stack
Add the models you use to see how this one compares.
Sign inAdditional source evidence
- Source model
- deepseek/deepseek-v4.1-flash-1017
- Source family
- deepseek-flash
- Last updated
- 2026-10-17
- Knowledge cutoff
- 2025-05
- Model context
- 1M
- Model output limit
- 384K
- Open weights
- Yes
- File attachments
- Yes
- Reasoning
- Yes
- Tool calling
- Yes
- Temperature control
- Yes
Downloadable weights
Published benchmarks
| Benchmark | Score | Evaluation details | Evidence |
|---|---|---|---|
| Humanity's Last Exam | 42.6 | full set · without tools | Source |
| Humanity's Last Exam | 44 | text-only subset · without tools | Source |
| Terminal-Bench | 42.3 | Claude Code · 4.0 | Source |
| DeepSWE | 77 | Mini-SWE-V2 · 1.1 | Source |
| ProgramBench | 29.1 | — | Source |
| NL2Repo | 68.2 | — | Source |
| AutomationBench | 60.5 | 1.0.6 | Source |
| Agents' Last Exam | 35.4 | — | Source |
| Apex Bench | 72.1 | 1.1 | Source |
| Chartography | 83 | with tools | Source |
| BabyVision | 92.5 | with tools | Source |
| ZeroBench | 51 · pass@5 | main · with tools | Source |
Complete source record (JSON)
{
"id": "deepseek/deepseek-v4.1-flash-1017",
"name": "DeepSeek V4.1 Flash 1017",
"description": "October 17 update of DeepSeek V4.1 Flash with improved reasoning, coding, agentic tasks, and visual understanding",
"family": "deepseek-flash",
"last_updated": "2026-10-17",
"knowledge": "2025-05",
"attachment": true,
"reasoning": true,
"tool_call": true,
"temperature": true,
"modalities": {
"input": [
"text",
"image"
],
"output": [
"text"
]
},
"limit": {
"context": 1000000,
"output": 384000
},
"benchmarks": [
{
"name": "Humanity's Last Exam",
"score": 42.6,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
"variant": "without tools",
"dataset": "full set"
},
{
"name": "Humanity's Last Exam",
"score": 44,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
"variant": "without tools",
"dataset": "text-only subset"
},
{
"name": "Terminal-Bench",
"score": 42.3,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
"harness": "Claude Code",
"version": "4.0"
},
{
"name": "DeepSWE",
"score": 77,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
"harness": "Mini-SWE-V2",
"version": "1.1"
},
{
"name": "ProgramBench",
"score": 29.1,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"
},
{
"name": "NL2Repo",
"score": 68.2,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"
},
{
"name": "AutomationBench",
"score": 60.5,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
"version": "1.0.6"
},
{
"name": "Agents' Last Exam",
"score": 35.4,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"
},
{
"name": "Apex Bench",
"score": 72.1,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
"version": "1.1"
},
{
"name": "Chartography",
"score": 83,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
"variant": "with tools"
},
{
"name": "BabyVision",
"score": 92.5,
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
"variant": "with tools"
},
{
"name": "ZeroBench",
"score": 51,
"source": "https://api-docs.deepseek.com/updates/",
"metric": "pass@5",
"variant": "with tools",
"dataset": "main"
}
],
"release_date": "2026-10-17",
"open_weights": true,
"license": "MIT",
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"
}
]
}Frequently asked questions
Running locally
At Q4 it is estimated to need 532.6 GB of VRAM, at Q8 933 GB and at FP16 1,755.9 GB, based on its 763B parameters. All figures are estimates, covering weights plus about 15% and 0.5 GB for a short context.
No, it does not fit on a 24 GB, 16 GB, 12 GB or 8 GB GPU at Q4, Q8 or FP16, according to estimates.
API and pricing
Input costs $0.006 per 1M tokens and output $0.50 per 1M tokens, tracked across 47 paid providers. Input prices range from $0.006 to $0.60 per 1M tokens.
Particle AI has the lowest blended price at a 3:1 input to output ratio, charging $0.025 per 1M input tokens and $0.25 per 1M output tokens. Relace has the lowest input price at $0.006 per 1M tokens.
Yes, cached input costs $0.005 per 1M tokens versus $0.006 uncached, a 17% discount.
Relace offers batch pricing at $0.006 per 1M input tokens and $0.50 per 1M output tokens, a 0% discount versus standard.
A request with 10,000 input tokens and 1,000 output tokens costs $0.0005 at the cheapest provider, Particle AI.
Of 47 endpoints with uptime data in the last day, 39 had 99% uptime or better. The best was 100.0%, by Alibaba, AtlasCloud, BaseTen and 22 more.
Privacy and data
Yes: Relace, Morph, Particle AI, Wafer, PrimeIntellect, OpenInference, DeepInfra, InferenceNet, Ionstream, Cloudflare, Sail Research, Cohere, Makora, Reka, Novita, DekaLLM, DigitalOcean, Fireworks, PrimaLabs, Modal, Crusoe, BaseTen, CoreWeave, Mistral, Near AI, Nebius, Parasail, SiliconFlow, Together and Venice.
Yes, DeepSeek V4.1 Flash 1017 has open weights, so it can be self-hosted for full control of your data.
Capabilities and specs
The context window is 1M tokens (1,048,576) and the maximum output is 944K tokens (943,718).
It accepts text and images. Audio, video and PDFs are not listed as input modalities.
Yes, tool calling (function calling) is supported.
Yes, structured output (JSON schema or JSON mode) is supported.
Yes, reasoning is supported, with effort levels max, high, low and none. The default effort is high and reasoning is not always on. 47 of 47 tracked providers expose reasoning, and a reasoning_effort setting is offered.
The knowledge cutoff is 2025-05.
Quality and benchmarks
It ranks 17 on Arena WebDev with a score of 1664 and 20 on Agent Arena with a score of 0. These are Arena's own ranks.
Yes, it is ahead of its predecessor DeepSeek V3.2 on ARC-AGI-2 and Arena WebDev, and behind or level on none.
Alternatives matched on model kind, blended price and release date include MiniMax M3.1 Flash by MiniMax at $0.1312 per 1M tokens, MiMo-V2.6-Flash by Xiaomi at $0.175 and Nex-N2.5-Pro by Nex AGI at $0.1188.
MiniMax M3.1 FlashDeepSeek V4.1 Flash 1017 vs MiniMax M3.1 FlashMiMo-V2.6-FlashDeepSeek V4.1 Flash 1017 vs MiMo-V2.6-FlashNex-N2.5-ProDeepSeek V4.1 Flash 1017 vs Nex-N2.5-Pro
Versions and lineage
Both have a 1M token context (1,048,576). The 1017 version costs $0.006 input and $0.50 output per 1M tokens versus $0.051 and $0.567 for DeepSeek V4.1 Flash, and was released October 17, 2026 versus September 10, 2026.
DeepSeek V4.1 FlashDeepSeek V4.1 Flash 1017 vs DeepSeek V4.1 Flash
Licensing and access
It has open weights under the MIT license.
Yes, the MIT license is a permissive license that allows commercial use.
The weights are available on the Hugging Face repository deepseek-ai/DeepSeek-V4.1-Flash-1017.
Basics
DeepSeek made it, and it was first listed on OpenRouter on October 17, 2026.
No, the listed API price is $0.006 per 1M input tokens and $0.50 per 1M output tokens, though the model has open weights.
Generated from Vertical Tabs data, updated November 11, 2026.
Discussion
No comments yet
Be the first to share your thoughts.