Where is the DeepSeek V4 Flash 0731 documentation?

The DeepSeek V4 Flash 0731 documentation on OpenRouter includes the Does DeepSeek V4 Have Vision? blog post, the Why Use OpenRouter for DeepSeek blog post, and the DeepSeek V4 Is Earning Agentic Token Share blog post. The quickstart covers API keys and your first request, and the API reference lists every parameter.

How much does DeepSeek V4 Flash 0731 cost?

DeepSeek V4 Flash 0731 costs $0.0434/M input tokens and $0.30/M output tokens, with separate rates for Cache Read at $0.035/M tokens.

What is the context length of DeepSeek V4 Flash 0731?

DeepSeek V4 Flash 0731 has a 1,048,576 token context window. It supports up to 1,048,576 completion tokens.

Does DeepSeek V4 Flash 0731 support tool calling and structured outputs?

Yes. DeepSeek V4 Flash 0731 accepts tools and tool_choice for function calling. It also supports structured outputs via a JSON schema in response_format.

Which providers serve DeepSeek V4 Flash 0731?

DeepSeek V4 Flash 0731 is served by 23 providers on OpenRouter: Wafer, Baidu Qianfan, DeepInfra, Relace, Reka AI, DigitalOcean, Baseten (fp32), Venice and 15 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 Flash 0731 released?

DeepSeek V4 Flash 0731 was released on July 31, 2026.

DeepSeek: DeepSeek V4 Flash 0731

deepseek/deepseek-v4-flash-0731

Model weights

Compare

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.

Modalities

In / out price

$0.0434 / $0.30per 1M

Context

1.0M

Released

Jul 31, 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


86% off	$0.44$0.05984	$1.32$0.1795	$0.014$0.001904	0.64s	102 tok/s	99.73%
	$0.07	$0.14	$0.014	0.43s	93 tok/s	99.97%
	$0.14	$0.28	$0.07	0.32s	177 tok/s	99.13%
	$0.0434	$0.30	$0.035	0.38s	117 tok/s	99.63%
	$0.119	$0.238	$0.0238	0.72s	164 tok/s	100.00%
	$0.06	$0.18	$0.015	0.43s	76 tok/s	100.00%
	$0.13	$0.26	$0.028	0.59s	99 tok/s	99.52%
	$0.13	$0.27	$0.05	0.42s	114 tok/s	99.97%
	$0.13	$0.495	$0.027	0.85s	74 tok/s	99.78%
	$0.14	$0.28	$0.05	0.54s	71 tok/s	98.79%
	$0.187	$0.561	$0.0187	1.75s	125 tok/s	100.00%
	$0.10	$0.42	$0.06	0.72s	144 tok/s	100.00%
	$0.15	$0.30	$0.042	0.58s	44 tok/s	99.56%
	$0.15	$0.50	$0.075	0.66s	69 tok/s	99.35%
	$0.15	$0.30	$0.042	1.21s	42 tok/s	99.73%
35% off	$0.44$0.286	$1.32$0.858	$0.014$0.0091	2.02s	63 tok/s	100.00%
	$0.22	$0.66	$0.028	0.98s	42 tok/s	95.39%
	$0.44	$1.32	$0.014	0.63s	90 tok/s	100.00%
	$0.12	$0.23	$0.115	0.45s	68 tok/s	99.44%
7% off	$0.44$0.4092	$1.32$1.228	$0.028$0.02604	1.00s	55 tok/s	99.94%
	$0.44	$1.32	$0.014	0.55s	121 tok/s	99.36%
70% off	$0.44$0.132	$1.32$0.396	$0.014$0.0042	4.85s	31 tok/s	100.00%
	$0.44	$1.32	$0.028	2.93s	55 tok/s	99.95%
	$0.176	$0.528	$0.0176	2.02s	84 tok/s	99.66%
	$0.13	$0.26	$0.028	0.33s	208 tok/s	100.00%
	$0.13	$0.26	$0.028	0.27s	146 tok/s	100.00%
	$0.44	$1.32	$0.044	--	--	57.14%

Input $0.05984Output $0.1795
86% off
102 tok/s99.73% uptime
Input $0.07Output $0.14
93 tok/s99.97% uptime
Input $0.14Output $0.28
177 tok/s99.13% uptime
Input $0.0434Output $0.30
117 tok/s99.63% uptime
Input $0.119Output $0.238
164 tok/s100.00% uptime
Input $0.06Output $0.18
76 tok/s100.00% uptime
Input $0.13Output $0.26
99 tok/s99.52% uptime
Input $0.13Output $0.27
114 tok/s99.97% uptime
Input $0.13Output $0.495
74 tok/s99.78% uptime
Input $0.14Output $0.28
71 tok/s98.79% uptime
Input $0.187Output $0.561
125 tok/s100.00% uptime
Input $0.10Output $0.42
144 tok/s100.00% uptime
Input $0.15Output $0.30
44 tok/s99.56% uptime
Input $0.15Output $0.50
69 tok/s99.35% uptime
Input $0.15Output $0.30
42 tok/s99.73% uptime
Input $0.286Output $0.858
35% off
63 tok/s100.00% uptime
Input $0.22Output $0.66
42 tok/s95.39% uptime
Input $0.44Output $1.32
90 tok/s100.00% uptime
Input $0.12Output $0.23
68 tok/s99.44% uptime
Input $0.4092Output $1.228
7% off
55 tok/s99.94% uptime
Input $0.44Output $1.32
121 tok/s99.36% uptime
Input $0.132Output $0.396
70% off
31 tok/s100.00% uptime
Input $0.44Output $1.32
55 tok/s99.95% uptime
Input $0.176Output $0.528
84 tok/s99.66% uptime
Input $0.13Output $0.26
208 tok/s100.00% uptime
Input $0.13Output $0.26
146 tok/s100.00% uptime
Input $0.44Output $1.32
-- tok/s57.14% 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

$0.04217/M tokens

Output

$0.296/M tokens

DeepSeek V4 Flash 0731 — Price history

Price history

No effective price data for this selection

$0.0123	$0.395	93.6%	7.2%
$0.0140	$0.178	79.1%	11.6%
$0.0319	$0.139	67.9%	8.2%
$0.0341	$0.178	57.4%	21.2%
$0.0369	$0.494	90.4%	1.3%
$0.0383	$0.238	84.7%	7.3%
$0.0384	$0.299	58.9%	18.5%
$0.0544	$0.300	88.5%	0.0%
$0.0566	$0.260	71.9%	2.1%
$0.0694	$0.270	75.8%	5.9%
$0.0716	$0.904	82.6%	0.1%
$0.0718	$0.420	70.5%	1.2%
$0.0751	$0.856	76.1%	1.5%
$0.0767	$0.279	70.3%	1.3%
$0.0831	$0.259	45.9%	0.3%
$0.0844	$0.259	44.6%	0.9%
$0.0852	$0.298	59.9%	0.5%
$0.0875	$0.279	75.0%	8.0%
$0.0962	$1.320	80.7%	0.3%
$0.0994	$1.229	81.0%	1.0%
$0.1049	$1.020	77.7%	0.0%
$0.1053	$0.660	59.7%	0.9%
$0.1076	$0.229	44.7%	0.4%
$0.1110	$0.500	51.9%	0.1%
$0.2192	$1.320	51.8%	0.0%
$0.3196	$1.319	29.2%	0.2%
—	—	—	—

Effective in $0.0123Effective out $0.395
Token share
7.2%
Effective in $0.0140Effective out $0.178
Token share
11.6%
Effective in $0.0319Effective out $0.139
Token share
8.2%
Effective in $0.0341Effective out $0.178
Token share
21.2%
Effective in $0.0369Effective out $0.494
Token share
1.3%
Effective in $0.0383Effective out $0.238
Token share
7.3%
Effective in $0.0384Effective out $0.299
Token share
18.5%
Effective in $0.0544Effective out $0.300
Token share
0.0%
Effective in $0.0566Effective out $0.260
Token share
2.1%
Effective in $0.0694Effective out $0.270
Token share
5.9%
Effective in $0.0716Effective out $0.904
Token share
0.1%
Effective in $0.0718Effective out $0.420
Token share
1.2%
Effective in $0.0751Effective out $0.856
Token share
1.5%
Effective in $0.0767Effective out $0.279
Token share
1.3%
Effective in $0.0831Effective out $0.259
Token share
0.3%
Effective in $0.0844Effective out $0.259
Token share
0.9%
Effective in $0.0852Effective out $0.298
Token share
0.5%
Effective in $0.0875Effective out $0.279
Token share
8.0%
Effective in $0.0962Effective out $1.320
Token share
0.3%
Effective in $0.0994Effective out $1.229
Token share
1.0%
Effective in $0.1049Effective out $1.020
Token share
0.0%
Effective in $0.1053Effective out $0.660
Token share
0.9%
Effective in $0.1076Effective out $0.229
Token share
0.4%
Effective in $0.1110Effective out $0.500
Token share
0.1%
Effective in $0.2192Effective out $1.320
Token share
0.0%
Effective in $0.3196Effective out $1.319
Token share
0.2%
Effective in —Effective out —
Token share —

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

208tok/s

P50, best across providers

Latency

0.27s

P50, best provider

Reasoning effortLatency / throughput

AutoExacto benchmarks

GPQA DiamondTAU-BenchMancer90.6%78.7%Morph90.4%78.0%Baidu Qianfan90.7%77.2%Together87.2%80.7%DeepInfra90.8%76.7%

+24 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)

99.87%

Availability over the last 3 days

Last 72 hours

Availability 99.87%

3 Days Ago2 Days AgoYesterdayNow

Availability over the last 24 hours

OpenRouter Availability

99.92%

Without Routing

94.96%

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 Flash 0731 (Artificial Analysis, Design Arena and Vals AI)
Source	Benchmark	Score
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) Coding Index	69.1
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) Agentic Index	41.0
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) GPQA Diamond	90.8%
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) HLE	38.6%
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) AA-LCR	79.7%
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) τ-Bench Banking	39.4%
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) GDPval-AA	46.8%
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) CritPt	16.6%
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) SciCode	50.3%
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) Terminal-Bench 4.0	12.1%
Artificial Analysis	DeepSeek V4 Flash 0731 (Max) Terminal-Bench 2.1	78.7%
Design Arena	DeepSeek-V4-Flash-0731 Models Arena 3D Elo	1188
Design Arena	DeepSeek-V4-Flash-0731 Models Arena Asciiart Elo	1092
Design Arena	DeepSeek-V4-Flash-0731 Models Arena Code Categories Elo	1219
Design Arena	DeepSeek-V4-Flash-0731 Models Arena Data Visualization Elo	1169
Design Arena	DeepSeek-V4-Flash-0731 Models Arena Game Development Elo	1198
Design Arena	DeepSeek-V4-Flash-0731 Models Arena SVG Elo	1170
Design Arena	DeepSeek-V4-Flash-0731 Models Arena UI Component Elo	1227
Design Arena	DeepSeek-V4-Flash-0731 Models Arena Website Elo	1230
Vals AI	DeepSeek V4 Flash 0731 BioMysteryBench	64.4%
Vals AI	DeepSeek V4 Flash 0731 Code Migration	38.6%
Vals AI	DeepSeek V4 Flash 0731 CorpFin v2	61.8%
Vals AI	DeepSeek V4 Flash 0731 CyberBench v1.1	72.1%
Vals AI	DeepSeek V4 Flash 0731 EMB	57.0%
Vals AI	DeepSeek V4 Flash 0731 Finance Agent (v2)	49.5%
Vals AI	DeepSeek V4 Flash 0731 GPQA Diamond	89.9%
Vals AI	DeepSeek V4 Flash 0731 Harvey's Legal Agent Benchmark v1	8.3%
Vals AI	DeepSeek V4 Flash 0731 IOI	32.7%
Vals AI	DeepSeek V4 Flash 0731 LiveCodeBench	87.3%
Vals AI	DeepSeek V4 Flash 0731 LegalBench	77.7%
Vals AI	DeepSeek V4 Flash 0731 Legal Research Bench	30.3%
Vals AI	DeepSeek V4 Flash 0731 MedCode	41.4%
Vals AI	DeepSeek V4 Flash 0731 MedScribe	80.4%
Vals AI	DeepSeek V4 Flash 0731 MMLU Pro	86.2%
Vals AI	DeepSeek V4 Flash 0731 ProgramBench	0.0%
Vals AI	DeepSeek V4 Flash 0731 ProofBench v1.1	56.0%
Vals AI	DeepSeek V4 Flash 0731 SkillsBench	50.7%
Vals AI	DeepSeek V4 Flash 0731 SWE-bench	88.8%
Vals AI	DeepSeek V4 Flash 0731 Tax Agent Bench	28.1%
Vals AI	DeepSeek V4 Flash 0731 TaxEval v2	70.7%
Vals AI	DeepSeek V4 Flash 0731 Terminal-Bench 4.0	18.7%
Vals AI	DeepSeek V4 Flash 0731 Terminal-Bench-Science	4.3%
Vals AI	DeepSeek V4 Flash 0731 Vals Index	48.0%
Vals AI	DeepSeek V4 Flash 0731 Vibe Code Bench v1.1	74.7%

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.

1.86Ttokens

LangChain

new

760Btokens

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

336Btokens

OpenClaw

OpenClaw is an open-source AI agent that connects to your messaging apps and takes real actions on your behalf, from running commands and browsing the web to managing files and sending emails.

293Btokens

Digen AI

new

182Btokens

Rest of OpenRouter

201B

Top 5 Apps

43.5B

Available through EU in-region routing

DeepSeek: DeepSeek V4 Flash 0731 is available through EU in-region routing. Requests sent to eu.openrouter.ai are decrypted in the EU and routed only to shared provider endpoints in Europe (the EU/EEA, the UK, and Switzerland).

Providers in Europe

Inceptron

Prices in Europe

Inceptron: $0.13/M tokens input, $0.495/M tokens output

Base URL

https://eu.openrouter.ai/api/v1

Plan

Requires Business or Enterprise. Self-serve on Business: no contract, no monthly minimum. OpenRouter Business for EU in-region routing

Fails closed: if no shared provider endpoint in Europe is available, the request errors instead of leaving the region. Requests served by your own BYOK key go to your deployment’s region, which you control. 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.

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.

Text1.0M context$0.038 / $0.50

DeepSeek V4.1 Flash 1017 (batch)

Text1.0M context$0.05 / $0.15

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

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.

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 (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 R1 0528 Qwen3 8B

DeepSeek-R1-0528 is a lightly upgraded release of DeepSeek R1 that taps more compute and smarter post-training tricks, pushing its reasoning and inference to the brink of flagship models like O3 and Gemini 2.5 Pro. It now tops math, programming, and logic leaderboards, showcasing a step-change in depth-of-thought. The distilled variant, DeepSeek-R1-0528-Qwen3-8B, transfers this chain-of-thought into an 8 B-parameter form, beating standard Qwen3 8B by +10 pp and tying the 235 B “thinking” giant on AIME 2024.

Text131K context

DeepSeek: DeepSeek V4 Flash 0731

Providers