Where is the DeepSeek V4.1 Flash documentation?

The DeepSeek V4.1 Flash documentation on OpenRouter includes the Self-Hosted LLM vs API: When Does Self-Hosting Pay Off blog post, the How to Benchmark LLMs blog post, the How to Stop an AI Agent Stuck in an Infinite Loop blog post, the LLM API Rate Limits: How to Fix OpenAI and Claude 429 Errors blog post, the How to Get Structured Output from Any LLM: JSON Schema, Tool Calls, or a Decision Model blog post, the Best OCR Model for Scanned Documents, Measured blog post, the Build an AI Code Reviewer with OpenRouter blog post, the Open-Weight vs Closed-Weight Models: What's the Difference? blog post, the How to Extract Tables from a PDF with an LLM blog post, and the Does DeepSeek V4 Have Vision? blog post. The quickstart covers API keys and your first request, and the API reference lists every parameter.

How much does DeepSeek V4.1 Flash cost?

DeepSeek V4.1 Flash costs $0.0507/M input tokens and $0.2028/M output tokens, with separate rates for Cache Read at $0.001014/M tokens.

What is the context length of DeepSeek V4.1 Flash?

DeepSeek V4.1 Flash has a 1,048,576 token context window. It supports up to 393,216 completion tokens.

Does DeepSeek V4.1 Flash support tool calling and structured outputs?

Yes. DeepSeek V4.1 Flash accepts tools and tool_choice for function calling. It also supports structured outputs via a JSON schema in response_format.

What inputs and outputs does DeepSeek V4.1 Flash support?

DeepSeek V4.1 Flash accepts text and images as input and returns text.

Which providers serve DeepSeek V4.1 Flash?

DeepSeek V4.1 Flash is served by 27 providers on OpenRouter: Baidu Qianfan, Wafer, Open Inference, inference.net, Alibaba Cloud Int., DeepInfra, River, Relace and 19 more. Requests are routed to the best available provider, with automatic failover to the others, and you can pin or exclude providers with provider routing.

When was DeepSeek V4.1 Flash released?

DeepSeek V4.1 Flash was released on September 10, 2026.

DeepSeek: DeepSeek V4.1 Flash (nitro)

deepseek/deepseek-v4.1-flash:nitro

Model weights

Compare

DeepSeek V4.1 Flash is a sparse mixture-of-experts model from DeepSeek, and the first built on the company's Causal Encoder-Decoder (CED) architecture. It activates 8B parameters on input and 16B on output from a 552B-parameter backbone, an asymmetric split that keeps per-token compute low relative to the model's total size. Image understanding is native to the architecture, with visual and text embeddings trained jointly from the start of pre-training rather than added afterward as in the earlier experimental V4 Flash Vision Exp.

It is suited for coding, terminal, and computer-use agents, along with long-horizon tasks that must run to completion across many steps and long-context analysis. Compressed KV caching cuts cache memory to roughly a quarter of the previous Flash generation, significantly reducing costs on agentic workloads. DeepSeek positions it as the cost-efficient tier of the V4.1 family and reports that it exceeds V4 Pro on performance, speed, and task completion time.

Modalities

In / out price

83% off

$0.0507 / $0.2028per 1M

Context

1.0M

Released

Sep 10, 2026

Providers

Different companies host the same model. OpenRouter routes your request to one of them based on the routing mode you pick: Balanced (price + speed), Nitro (fastest), or Floor (cheapest).

Latency / throughput

Provider	Input /M	Output /M	Cache read /M	Latency	Throughput	Uptime
	$0.30	$1.20	$0.006	0.19s	280 tok/s	100.00%
	$0.30	$1.20	$0.007	0.42s	210 tok/s	99.81%
	$0.051	$0.567	$0.018	0.26s	202 tok/s	100.00%
	$0.19	$0.65	$0.012	0.22s	190 tok/s	100.00%
	$0.30	$1.20	$0.007	0.26s	187 tok/s	99.81%
	$0.45	$1.80	$0.009	0.28s	183 tok/s	100.00%
	$0.30	$1.20	$0.0075	0.45s	178 tok/s	99.50%
	$0.30	$1.20	$0.006	0.36s	174 tok/s	100.00%
	$0.29	$1.20	$0.007	0.54s	171 tok/s	100.00%
	$0.30	$1.20	$0.006	0.48s	170 tok/s	99.41%
Fast	$0.45	$1.80	$0.009	0.38s	161 tok/s	100.00%
	$0.30	$1.20	$0.03	0.41s	160 tok/s	100.00%
	$0.225	$0.90	$0.006	0.61s	158 tok/s	100.00%
	$0.195	$0.66	$0.003	0.28s	146 tok/s	100.00%
	$0.12	$0.48	$0.0024	0.65s	142 tok/s	99.94%
	$0.26	$1.17	$0.0058	1.07s	123 tok/s	100.00%
	$0.12	$0.50	$0.003	0.47s	117 tok/s	100.00%
	$0.13	$0.65	$0.002	0.62s	116 tok/s	100.00%
	$0.21$0.105	$0.84$0.42	$0.0042$0.0021	0.76s	109 tok/s	100.00%
35% off	$0.30$0.195	$1.20$0.78	$0.006$0.0039	0.94s	107 tok/s	99.64%
30% off	$0.20$0.14	$0.60$0.42	$0.006$0.0042	0.58s	103 tok/s	99.74%
	$0.30	$1.20	$0.006	1.58s	102 tok/s	99.82%
	$0.103	$0.60	$0.008	0.27s	101 tok/s	100.00%
Priority30% off	$0.30$0.21	$0.90$0.63	$0.009$0.0063	0.48s	99 tok/s	99.71%
83% off	$0.30$0.0507	$1.20$0.2028	$0.006$0.001014	1.61s	93 tok/s	99.89%
	$0.103	$0.60	$0.008	0.30s	93 tok/s	99.97%
	$0.135	$0.54	$0.0135	0.66s	90 tok/s	99.79%
	$0.21	$0.84	$0.0042	1.20s	85 tok/s	98.36%
	$0.30	$1.20	$0.03	0.62s	82 tok/s	99.45%
51% off	$0.30$0.147	$1.20$0.588	$0.006$0.00294	1.17s	80 tok/s	99.77%
	$0.2125	$0.75	$0.05	0.70s	79 tok/s	99.84%
	$0.07	$0.80	$0.006	1.54s	62 tok/s	100.00%
40% off	$0.30$0.18	$1.20$0.72	$0.006$0.0036	1.81s	46 tok/s	99.66%
20% off	$0.15$0.12	$0.80$0.64	$0.01$0.008	1.60s	41 tok/s	98.75%
Not used in Nitro routing:Why these endpoints are not used
Ultrafast	$3.00	$12.00	$0.06	1.49s	260 tok/s	100.00%
Flex30% off	$0.16$0.112	$0.48$0.336	$0.0048$0.00336	1.62s	140 tok/s	96.52%

Input $0.30Output $1.20
280 tok/s100.00% uptime
Input $0.30Output $1.20
210 tok/s99.81% uptime
Input $0.051Output $0.567
202 tok/s100.00% uptime
Input $0.19Output $0.65
190 tok/s100.00% uptime
Input $0.30Output $1.20
187 tok/s99.81% uptime
Input $0.45Output $1.80
183 tok/s100.00% uptime
Input $0.30Output $1.20
178 tok/s99.50% uptime
Input $0.30Output $1.20
174 tok/s100.00% uptime
Input $0.29Output $1.20
171 tok/s100.00% uptime
Input $0.30Output $1.20
170 tok/s99.41% uptime
Input $0.45Output $1.80
Fast
161 tok/s100.00% uptime
Input $0.30Output $1.20
160 tok/s100.00% uptime
Input $0.225Output $0.90
158 tok/s100.00% uptime
Input $0.195Output $0.66
146 tok/s100.00% uptime
Input $0.12Output $0.48
142 tok/s99.94% uptime
Input $0.26Output $1.17
123 tok/s100.00% uptime
Input $0.12Output $0.50
117 tok/s100.00% uptime
Input $0.13Output $0.65
116 tok/s100.00% uptime
Input $0.105Output $0.42
109 tok/s100.00% uptime
Input $0.195Output $0.78
35% off
107 tok/s99.64% uptime
Input $0.14Output $0.42
30% off
103 tok/s99.74% uptime
Input $0.30Output $1.20
102 tok/s99.82% uptime
Input $0.103Output $0.60
101 tok/s100.00% uptime
Input $0.21Output $0.63
Priority
1 more label
99 tok/s99.71% uptime
Input $0.0507Output $0.2028
83% off
93 tok/s99.89% uptime
Input $0.103Output $0.60
93 tok/s99.97% uptime
Input $0.135Output $0.54
90 tok/s99.79% uptime
Input $0.21Output $0.84
85 tok/s98.36% uptime
Input $0.30Output $1.20
82 tok/s99.45% uptime
Input $0.147Output $0.588
51% off
80 tok/s99.77% uptime
Input $0.2125Output $0.75
79 tok/s99.84% uptime
Input $0.07Output $0.80
62 tok/s100.00% uptime
Input $0.18Output $0.72
40% off
46 tok/s99.66% uptime
Input $0.12Output $0.64
20% off
41 tok/s98.75% uptime
Not used in Nitro routing:Why these endpoints are not used
Input $3.00Output $12.00
Ultrafast
260 tok/s100.00% uptime
Input $0.112Output $0.336
Flex
1 more label
140 tok/s96.52% uptime

Pricing

The average price customers actually pay for this model, next to the prices providers post. Caching and discounts mean the price actually paid is often well below the listed one.

Weighted average

Input

$0.02877/M tokens

Output

$0.6262/M tokens

DeepSeek V4.1 Flash — Price history

Price history

No effective price data for this selection

		$0.00431	$0.583	99.0%	2.7%
		$0.00820	$0.202	85.5%	9.5%
		$0.00853	$0.800	96.0%	10.5%
		$0.01267	$0.500	91.7%	2.9%
		$0.01318	$0.479	90.8%	6.3%
		$0.01336	$0.420	93.3%	14.5%
		$0.01713	$0.600	90.4%	1.9%
		$0.01925	$0.765	91.6%	4.6%
		$0.01986	$0.659	91.2%	4.7%
		$0.01989	$0.839	92.4%	1.6%
		$0.01998	$0.899	93.6%	4.9%
		$0.02097	$0.567	90.9%	8.7%
		$0.02140	$0.649	84.8%	3.3%
		$0.02179	$0.599	85.4%	1.4%
		$0.02600	$1.200	93.2%	1.0%
	Priority	$0.02661	$0.630	90.0%	0.5%
		$0.02832	$1.199	92.5%	2.0%
		$0.02862	$1.170	91.0%	0.6%
		$0.02917	$0.637	80.9%	0.1%
		$0.02927	$1.200	92.4%	2.4%
		$0.03053	$0.649	89.6%	3.6%
		$0.03260	$1.200	90.9%	0.6%
		$0.03471	$1.199	90.5%	0.1%
		$0.03759	$1.200	89.7%	0.9%
		$0.03993	$0.721	79.4%	0.1%
		$0.04887	$1.199	93.0%	0.1%
		$0.06138	$1.799	88.1%	0.4%
		$0.06997	$0.533	53.4%	0.2%
		$0.07293	$0.746	85.6%	0.0%
		$0.07331	$1.200	77.1%	0.1%
		$0.07354	$0.800	63.9%	0.2%
	Fast	$0.10015	$1.799	79.3%	0.1%
		$0.14062	$1.200	59.0%	0.1%
		$0.14718	$1.199	52.0%	8.8%
Not used in Nitro routing:Why these endpoints are not used
	Flex	$0.03160	$0.336	74.0%	0.5%
	Ultrafast	$0.40695	$12.000	88.2%	0.0%

Effective in $0.00431Effective out $0.583
Token share
2.7%
Effective in $0.00820Effective out $0.202
Token share
9.5%
Effective in $0.00853Effective out $0.800
Token share
10.5%
Effective in $0.01267Effective out $0.500
Token share
2.9%
Effective in $0.01318Effective out $0.479
Token share
6.3%
Effective in $0.01336Effective out $0.420
Token share
14.5%
Effective in $0.01713Effective out $0.600
Token share
1.9%
Effective in $0.01925Effective out $0.765
Token share
4.6%
Effective in $0.01986Effective out $0.659
Token share
4.7%
Effective in $0.01989Effective out $0.839
Token share
1.6%
Effective in $0.01998Effective out $0.899
Token share
4.9%
Effective in $0.02097Effective out $0.567
Token share
8.7%
Effective in $0.02140Effective out $0.649
Token share
3.3%
Effective in $0.02179Effective out $0.599
Token share
1.4%
Effective in $0.02600Effective out $1.200
Token share
1.0%
Effective in $0.02661Effective out $0.630
PriorityToken share
0.5%
Effective in $0.02832Effective out $1.199
Token share
2.0%
Effective in $0.02862Effective out $1.170
Token share
0.6%
Effective in $0.02917Effective out $0.637
Token share
0.1%
Effective in $0.02927Effective out $1.200
Token share
2.4%
Effective in $0.03053Effective out $0.649
Token share
3.6%
Effective in $0.03260Effective out $1.200
Token share
0.6%
Effective in $0.03471Effective out $1.199
Token share
0.1%
Effective in $0.03759Effective out $1.200
Token share
0.9%
Effective in $0.03993Effective out $0.721
Token share
0.1%
Effective in $0.04887Effective out $1.199
Token share
0.1%
Effective in $0.06138Effective out $1.799
Token share
0.4%
Effective in $0.06997Effective out $0.533
Token share
0.2%
Effective in $0.07293Effective out $0.746
Token share
0.0%
Effective in $0.07331Effective out $1.200
Token share
0.1%
Effective in $0.07354Effective out $0.800
Token share
0.2%
Effective in $0.10015Effective out $1.799
FastToken share
0.1%
Effective in $0.14062Effective out $1.200
Token share
0.1%
Effective in $0.14718Effective out $1.199
Token share
8.8%
Not used in Nitro routing:Why these endpoints are not used
Effective in $0.03160Effective out $0.336
FlexToken share
0.5%
Effective in $0.40695Effective out $12.000
UltrafastToken share
0.0%

Performance

Throughput is how fast the model writes (tokens per second — higher is better). Latency is total round-trip time (lower is better). TTFT is time-to-first-token — how long before you see anything appear (lower is better).

Throughput

280tok/s

P50, best across providers

Latency

0.19s

P50, best provider

Reasoning effortLatency / throughput

AutoExacto benchmarks

GPQA DiamondTAU-BenchBaseten (US)89.7%--Cloudflare87.7%--DeepInfra91.2%75.9%StreamLake90.9%75.9%DigitalOcean90.6%76.2%Baseten90.1%73.7%Baseten89.4%71.4%

+32 more providers

Uptime

Uptime is the percentage of the past 3 days that at least one provider was responding to requests. Availability is the percentage of time that inference was successfully served. OpenRouter continuously monitors and uses the next-best provider when one returns an error.

Uptime (3d)

100.00%

Availability (3d)

99.85%

Availability over the last 3 days

Last 72 hours

Availability 99.85%

3 Days Ago2 Days AgoYesterdayNow

Availability over the last 24 hours

OpenRouter Availability

99.94%

Without Routing

99.31%

When an error occurs in an upstream provider, we can recover by routing to another healthy provider, if your request filters allow it. You can access per-provider uptime data programmatically through the Endpoints API. Learn more about our load balancing and provider routing options.

Benchmarks

Scores on standardized evaluations and this model’s own outputs on shared benchmark prompts. Higher percentages are better — and each rank shows where this model lands among the models evaluated on that source.

Benchmark score summary for DeepSeek: DeepSeek V4.1 Flash (Artificial Analysis and Vals AI)
Source	Benchmark	Score
Artificial Analysis	DeepSeek V4.1 Flash (Non-reasoning) HLE	10.8%
Artificial Analysis	DeepSeek V4.1 Flash (Non-reasoning) AA-LCR	55.3%
Artificial Analysis	DeepSeek V4.1 Flash (Non-reasoning) GDPval-AA	42.4%
Artificial Analysis	DeepSeek V4.1 Flash (Non-reasoning) CritPt	0.3%
Artificial Analysis	DeepSeek V4.1 Flash (Non-reasoning) SciCode	35.5%
Artificial Analysis	DeepSeek V4.1 Flash (Non-reasoning) Terminal-Bench 4.0	5.6%
Vals AI	DeepSeek V4.1 Flash Access to Justice Benchmark	72.9%
Vals AI	DeepSeek V4.1 Flash BioMysteryBench	67.8%
Vals AI	DeepSeek V4.1 Flash Code Migration v1.1	58.9%
Vals AI	DeepSeek V4.1 Flash CyberBench v1.1	73.7%
Vals AI	DeepSeek V4.1 Flash EMB v1.1	67.8%
Vals AI	DeepSeek V4.1 Flash Finance Agent v2.1	65.4%
Vals AI	DeepSeek V4.1 Flash Vals Finance Index	57.1%
Vals AI	DeepSeek V4.1 Flash Financial Coaching Bench	72.8%
Vals AI	DeepSeek V4.1 Flash Harvey's Legal Agent Benchmark v1	6.7%
Vals AI	DeepSeek V4.1 Flash Immigration	65.6%
Vals AI	DeepSeek V4.1 Flash Vals Inference Provider Index	66.6%
Vals AI	DeepSeek V4.1 Flash IOI	40.3%
Vals AI	DeepSeek V4.1 Flash LegalBench	83.3%
Vals AI	DeepSeek V4.1 Flash Legal Research Bench	41.3%
Vals AI	DeepSeek V4.1 Flash MedCode	41.2%
Vals AI	DeepSeek V4.1 Flash MedScribe	85.5%
Vals AI	DeepSeek V4.1 Flash MysteryMechanism	21.2%
Vals AI	DeepSeek V4.1 Flash ProgramBench	0.5%
Vals AI	DeepSeek V4.1 Flash ProofBench v1.1	54.0%
Vals AI	DeepSeek V4.1 Flash Public Benefits Bench v1.1	64.3%
Vals AI	DeepSeek V4.1 Flash SAGE	85.4%
Vals AI	DeepSeek V4.1 Flash Vals Science Index	29.6%
Vals AI	DeepSeek V4.1 Flash SkillsBench	69.8%
Vals AI	DeepSeek V4.1 Flash SRE Bench	0.8%
Vals AI	DeepSeek V4.1 Flash SWE-bench Multimodal	46.5%
Vals AI	DeepSeek V4.1 Flash Tax Agent Bench	26.6%
Vals AI	DeepSeek V4.1 Flash Terminal-Bench 4.0	19.7%
Vals AI	DeepSeek V4.1 Flash Terminal-Bench-Science	0.0%
Vals AI	DeepSeek V4.1 Flash Vals Index	51.3%
Vals AI	DeepSeek V4.1 Flash Vibe Code Bench 1-100	16.4%
Vals AI	DeepSeek V4.1 Flash Vibe Code Bench v1.1	84.7%

Apps

Public apps that send the most traffic to this model. Good signal for what real production workloads look like — and a hint at which use cases this model is best suited for.

Hermes Agent

Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents.

9.57Ttokens

Claude Code

Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans and executes changes across files, runs tests, and iterates on failures, all from natural language prompts.

4.05Ttokens

There are many coding agents, but this one is yours.

3.28Ttokens

DeepSeek Harness

new

2.14Ttokens

OpenHands

AI coding agent that can run commands, browse the web, call APIs

2.05Ttokens

Rest of OpenRouter

898B

Top 5 Apps

460B

DeepSeek: DeepSeek Flash versions

Not currently available through EU in-region routing. In-region routing docs

More models from DeepSeek

DeepSeek V4.1 Flash 1017

DeepSeek V4.1 Flash 1017 is the October 17, 2026 snapshot of DeepSeek V4.1 Flash, served by DeepSeek under the deepseek-flash API model name. V4.1 Flash is a sparse mixture-of-experts model from DeepSeek, and the first built on the company's Causal Encoder-Decoder (CED) architecture. It activates 8B parameters on input and 16B on output from a 552B-parameter backbone, an asymmetric split that keeps per-token compute low relative to the model's total size. Image understanding is native to the architecture, with visual and text embeddings trained jointly from the start of pre-training rather than added afterward as in the earlier experimental V4 Flash Vision Exp.

Text1.0M context$0.02 / $0.108

DeepSeek V4.1 Flash 1017 (batch)

Text1.0M context$0.036 / $0.319

DeepSeek Pro Latest

This model always redirects to the latest model in the DeepSeek Pro family.

Text1.0M context

DeepSeek Flash Latest

This model always redirects to the latest model in the DeepSeek Flash family.

Text1.0M context

DeepSeek V4.1 Flash

Text1.0M context$0.0507 / $0.2028

DeepSeek V4.1 Flash (batch)

Text1.0M context$0.112 / $0.336

DeepSeek V4 Flash Vision Exp

DeepSeek V4 Flash Vision Exp is an experimental vision-enabled version of DeepSeek V4 Flash 0731 from DeepSeek, adding image understanding while matching the base model on text capabilities including agents, reasoning, and world knowledge. It is a sparse mixture-of-experts model with 13B active parameters out of 284B total.

It is suited for document and chart understanding, visual question answering, and multimodal agent workflows that interleave text and images.

Text1.0M context$0.2156 / $0.6468

DeepSeek V4 Pro 0813

DeepSeek V4 Pro 0813 is a large-scale mixture-of-experts model from DeepSeek. This is the GA release of DeepSeek V4 Pro.

Text1.0M context$0.0999 / $8

DeepSeek V4 Flash Latest

This model always redirects to the latest model in the DeepSeek V4 Flash family.

Text1.0M context

DeepSeek V4 Flash 0731

DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows. This is the GA release of DeepSeek V4 Flash.

Text1.0M context$0.0434 / $0.30

DeepSeek V4 Flash 0731 (batch)

Text1.0M context$0.048 / $0.144

DeepSeek V4 Pro 0423

DeepSeek V4 Pro is a large-scale Mixture-of-Experts model from DeepSeek with 1.6T total parameters and 49B activated parameters, supporting a 1M-token context window. It is designed for advanced reasoning, coding, and long-horizon agent workflows, with strong performance across knowledge, math, and software engineering benchmarks.

Built on the same architecture as DeepSeek V4 Flash, it introduces a hybrid attention system for efficient long-context processing. Reasoning efforts high and xhigh are supported; xhigh maps to max reasoning. It is well suited for complex workloads such as full-codebase analysis, multi-step automation, and large-scale information synthesis, where both capability and efficiency are critical

Text1.0M context$0.2079 / $0.4157

DeepSeek V4 Flash 0423

DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window. It is designed for fast inference and high-throughput workloads, while maintaining strong reasoning and coding performance.

The model includes hybrid attention for efficient long-context processing. Reasoning efforts high and xhigh are supported; xhigh maps to max reasoning. It is well suited for applications such as coding assistants, chat systems, and agent workflows where responsiveness and cost efficiency are important.

Text1.0M context$0.014 / $0.45

DeepSeek V3.2 Speciale

DeepSeek-V3.2-Speciale is a high-compute variant of DeepSeek-V3.2 optimized for maximum reasoning and agentic performance. It builds on DeepSeek Sparse Attention (DSA) for efficient long-context processing, then scales post-training reinforcement learning to push capability beyond the base model. Reported evaluations place Speciale ahead of GPT-5 on difficult reasoning workloads, with proficiency comparable to Gemini-3.0-Pro, while retaining strong coding and tool-use reliability. Like V3.2, it benefits from a large-scale agentic task synthesis pipeline that improves compliance and generalization in interactive environments.

Text131K context

DeepSeek V3.2

DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism that reduces training and inference cost while preserving quality in long-context scenarios. A scalable reinforcement learning post-training framework further improves reasoning, with reported performance in the GPT-5 class, and the model has demonstrated gold-medal results on the 2025 IMO and IOI. V3.2 also uses a large-scale agentic task synthesis pipeline to better integrate reasoning into tool-use settings, boosting compliance and generalization in interactive environments.

Users can control the reasoning behaviour with the reasoning enabled boolean. Learn more in our docs

Text164K context$0.2088 / $0.3096

DeepSeek V3.2 Exp

DeepSeek-V3.2-Exp is an experimental large language model released by DeepSeek as an intermediate step between V3.1 and future architectures. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism designed to improve training and inference efficiency in long-context scenarios while maintaining output quality. Users can control the reasoning behaviour with the reasoning enabled boolean. Learn more in our docs

The model was trained under conditions aligned with V3.1-Terminus to enable direct comparison. Benchmarking shows performance roughly on par with V3.1 across reasoning, coding, and agentic tool-use tasks, with minor tradeoffs and gains depending on the domain. This release focuses on validating architectural optimizations for extended context lengths rather than advancing raw task accuracy, making it primarily a research-oriented model for exploring efficient transformer designs.

Text164K context$0.27 / $0.41

DeepSeek V3.1 Terminus

DeepSeek-V3.1 Terminus is an update to DeepSeek V3.1 that maintains the model's original capabilities while addressing issues reported by users, including language consistency and agent capabilities, further optimizing the model's performance in coding and search agents. It is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes. It extends the DeepSeek-V3 base with a two-phase long-context training process, reaching up to 128K tokens, and uses FP8 microscaling for efficient inference. Users can control the reasoning behaviour with the reasoning enabled boolean. Learn more in our docs

The model improves tool use, code generation, and reasoning efficiency, achieving performance comparable to DeepSeek-R1 on difficult benchmarks while responding more quickly. It supports structured tool calling, code agents, and search agents, making it suitable for research, coding, and agentic workflows.

Text164K context$0.27 / $1

DeepSeek V3.1

DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It extends the DeepSeek-V3 base with a two-phase long-context training process, reaching up to 128K tokens, and uses FP8 microscaling for efficient inference. Users can control the reasoning behaviour with the reasoning enabled boolean. Learn more in our docs

It succeeds the DeepSeek V3-0324 model and performs well on a variety of tasks.

Text164K context$0.25 / $0.95

DeepSeek V3.1 Base

This is a base model, trained only for raw next-token prediction. Unlike instruct/chat models, it has not been fine-tuned to follow user instructions. Prompts need to be written more like training text or examples rather than simple requests (e.g., “Translate the following sentence…” instead of just “Translate this”).

DeepSeek-V3.1 Base is a 671B parameter open Mixture-of-Experts (MoE) language model with 37B active parameters per forward pass and a context length of 128K tokens. Trained on 14.8T tokens using FP8 mixed precision, it achieves high training efficiency and stability, with strong performance across language, reasoning, math, and coding tasks.

Text164K context

R1 Distill Qwen 7B

DeepSeek-R1-Distill-Qwen-7B is a 7 billion parameter dense language model distilled from DeepSeek-R1, leveraging reinforcement learning-enhanced reasoning data generated by DeepSeek's larger models. The distillation process transfers advanced reasoning, math, and code capabilities into a smaller, more efficient model architecture based on Qwen2.5-Math-7B. This model demonstrates strong performance across mathematical benchmarks (92.8% pass@1 on MATH-500), coding tasks (Codeforces rating 1189), and general reasoning (49.1% pass@1 on GPQA Diamond), achieving competitive accuracy relative to larger models while maintaining smaller inference costs.

Text131K context

DeepSeek: DeepSeek V4.1 Flash (nitro)

Providers