DeepSeek-V3

DeepSeek-V3 belongs to the DeepSeek model family that has shaped the project's path from an early general-purpose chat model to later reasoning-focused releases. The BentoML guide positions DeepSeek-V3 as the foundational generation that preceded specialized reasoning variants, framing it as the base architecture from which subsequent versions such as V3-0324, V3.1, V3.2, and ultimately V4 were developed and improved upon. This lineage matters because it situates V3 as a broad text-to-text model designed to handle open-ended conversation, instruction following, and general knowledge queries, rather than as a narrowly specialized reasoning system.

Later DeepSeek releases built directly on the V3 foundation, with the team adding architectural refinements and reasoning capabilities that reportedly closed the gap with frontier systems on reasoning benchmarks. DeepSeek-V3.2-Speciale, the high-compute variant of the V3 line, has been cited as outperforming GPT-5 on reasoning tasks and rivaling Gemini-3.0-Pro, suggesting that the underlying V3-style architecture scales effectively when paired with stronger training and inference compute. For practitioners, DeepSeek-V3 is best understood as a capable general-purpose text model from an actively evolving family, suitable for applications that benefit from very large context windows and that may later migrate to newer reasoning-optimized variants in the same lineage.

Qiniudeepseek-v3deepseekdeprecated

Open provider page Provider docs

Quick Info

Provider: Qiniu
Model key: deepseek-v3
Release date: Dec 26, 2024
Last updated: Dec 26, 2024
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.289855
Output token cost: $1.15942

Limits

Output tokens: 16,000 tokens
Context window: 128,000 tokens

Transparent token rates

Compare DeepSeek-V3 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about DeepSeek-V3

iFlow

Official sourceAnnouncement

DeepSeek announced DeepSeek-V3 on December 25, 2024, describing it as the "biggest leap forward yet" with 671B MoE parameters, 37B activated parameters, and training on 14.8T high-quality tokens. Generation throughput reached 60 tokens per second, roughly 3x faster than V2, while keeping API compatibility and shipping fully open-source models and papers. The launch also disclosed API pricing for DeepSeek-V3: through February 8, 2025, rates matched V2, and thereafter input was priced at $0.27/M tokens (cache miss) and $0.07/M tokens (cache hit), with output at $1.10/M tokens. The announcement highlighted open-source release, with multimodal support and further DeepSeek ecosystem features teased as future work.

Qiniu

CoveragePreview

DeepSeek says both models are more efficient and performant than DeepSeek V3.2 due to architectural improvements, and have almost "closed the gap" with current leading models, both open and closed, on reasoning benchmarks.

iFlow

Coverage

The DeepSeek-V3 Technical Report (arXiv 2412.19437), submitted by DeepSeek-AI on December 27, 2024 and revised February 18, 2025, provides the canonical technical reference for DeepSeek-V3. It documents the architecture and training methodology behind a 671B-parameter MoE model with 37B activated parameters, trained on 14.8T tokens. The paper details core techniques including Multi-head Latent Attention (MLA), DeepSeekMoE, FP8 mixed-precision training, and multi-token prediction, with a large multi-author team led by DeepSeek-AI. It serves as the authoritative technical companion to the V3 release, supporting the model's open-source research artifacts alongside the GitHub repository.

Qiniu

Coverage

DeepSeek released DeepSeek-V3.2, a family of open-source reasoning and agentic AI models. The high compute version, DeepSeek-V3.2-Speciale, performs better than GPT-5 and comparably to Gemini-3.0-Pro

Sulat also reports on AI and tech beyond the catalog.Read news.sulat.com