DeepSeek Researchers Introduce DeepSeek-V3.2 and DeepSeek-V3.2-Speciale for Lengthy Context Reasoning and Agentic Workloads

How do you get GPT-5-level reasoning on actual long-context, tool-using workloads with out paying the quadratic consideration and GPU price that normally makes these methods impractical? DeepSeek analysis introduces DeepSeek-V3.2 and DeepSeek-V3.2-Speciale. They’re reasoning-first fashions constructed for brokers and targets prime quality reasoning, lengthy context and agent workflows, with open weights and manufacturing APIs. The fashions mix DeepSeek Sparse Consideration (DSA), a scaled GRPO reinforcement studying stack and an agent native software protocol, and report efficiency similar to GPT 5, with DeepSeek-V3.2-Speciale reaching Gemini 3.0 Professional degree reasoning on public benchmarks and competitions.

Sparse Consideration with Close to Linear Lengthy Context Value

Each DeepSeek-V3.2 and DeepSeek-V3.2-Speciale use the DeepSeek-V3 Combination of Consultants transformer with about 671B whole parameters and 37B lively parameters per token, inherited from V3.1 Terminus. The one structural change is DeepSeek Sparse Consideration, launched by means of continued pre-training.

DeepSeek Sparse Consideration splits consideration into 2 elements. A lightning indexer runs a small variety of low precision heads over all token pairs and produces relevance scores. A positive grained selector retains the top-k-key worth positions per question, and the principle consideration path runs Multi-Question-Consideration and Multi-Head-Latent-Consideration on this sparse set.

This modifications the dominant complexity from O(L²) to O(kL), the place L is sequence size and okay is the variety of chosen tokens and far smaller than L. Based mostly on the benchmarks, DeepSeek-V3.2 matches the dense Terminus baseline on accuracy whereas decreasing lengthy context inference price by about 50 p.c, with quicker throughput and decrease reminiscence use on H800 class {hardware} and on vLLM and SGLang backends.

Continued Pre Coaching for DeepSeek Sparse Consideration

DeepSeek Sparse Consideration (DSA) is launched by continued pre-training on high of DeepSeek-V3.2 Terminus. Within the dense heat up stage, dense consideration stays lively, all spine parameters are frozen and solely the lightning indexer is skilled with a Kullback Leibler loss to match the dense consideration distribution on 128K context sequences. This stage makes use of a small variety of steps and about 2B tokens, sufficient for the indexer to be taught helpful scores.

Within the sparse stage, the selector retains 2048 key-value entries per question, the spine is unfrozen and the mannequin continues coaching on about 944B tokens. Gradients for the indexer nonetheless come solely from the alignment loss with dense consideration on the chosen positions. This schedule makes DeepSeek Sparse Consideration (DSA) behave as a drop in substitute for dense consideration with related high quality and decrease lengthy context price.

GRPO with greater than 10 P.c RL Compute

On high of the sparse structure, DeepSeek-V3.2 makes use of Group Relative Coverage Optimization (GRPO) as the principle reinforcement studying technique. The analysis group state that submit coaching reinforcement studying RL compute exceeds 10 p.c of pre coaching compute.

RL is organized round specialist domains. The analysis group trains devoted runs for arithmetic, aggressive programming, common logical reasoning, searching and agent duties and security, then distills these specialists into the shared 685B parameter base for DeepSeek-V3.2 and DeepSeek-V3.2-Speciale. GRPO is carried out with an unbiased KL estimator, off coverage sequence masking and mechanisms that preserve Combination of Consultants (MoE) routing and sampling masks constant between coaching and sampling.

Agent Knowledge, Pondering Mode and Instrument Protocol

DeepSeek analysis group builds a big artificial agent dataset by producing greater than 1,800 environments and greater than 85,000 duties throughout code brokers, search brokers, common instruments and code interpreter setups. Duties are constructed to be onerous to resolve and simple to confirm, and are used as RL targets along with actual coding and search traces.

At inference time, DeepSeek-V3.2 introduces specific considering and non considering modes. The deepseek-reasoner endpoint exposes considering mode by default, the place the mannequin produces an inside chain of thought earlier than the ultimate reply. The considering with instruments information describes how reasoning content material is stored throughout software calls and cleared when a brand new consumer message arrives, and the way software calls and power outcomes keep within the context even when reasoning textual content is trimmed for price range.

The chat template is up to date round this habits. The DeepSeek-V3.2 Speciale repository ships Python encoder and decoder helpers as an alternative of a Jinja template. Messages can carry a reasoning_content discipline alongside content material, managed by a considering parameter. A developer position is reserved for search brokers and isn’t accepted normally chat flows by the official API, which protects this channel from unintended misuse.

Benchmarks, Competitions And Open Artifacts

On commonplace reasoning and coding benchmarks, DeepSeek-V3.2 and particularly DeepSeek-V3.2 Speciale are reported as similar to GPT-5 and near Gemini-3.0 Professional on suites resembling AIME 2025, HMMT 2025, GPQA and LiveCodeBench, with improved price effectivity on lengthy context workloads.

For formal competitions, DeepSeek analysis group states that DeepSeek-V3.2 Speciale achieves gold medal degree efficiency on the Worldwide Mathematical Olympiad 2025, the Chinese language Mathematical Olympiad 2025 and the Worldwide Olympiad in Informatics 2025, and aggressive gold medal degree efficiency on the ICPC World Finals 2025.

Key Takeaways

1. DeepSeek-V3.2 provides DeepSeek Sparse Consideration, which brings close to linear O(kL) consideration price and delivers round 50% decrease lengthy context API price in comparison with earlier dense DeepSeek fashions, whereas maintaining high quality just like DeepSeek-V3.1 Terminus.
2. The mannequin household retains the 671B parameter MoE spine with 37B lively parameters per token and exposes a full 128K context window in manufacturing APIs, which makes lengthy paperwork, multi step chains and enormous software traces sensible somewhat than a lab solely characteristic.
3. Publish coaching makes use of Group Relative Coverage Optimization (GRPO) with a compute price range that’s greater than 10 p.c of pre-training, centered on math, code, common reasoning, searching or agent workloads and security, together with contest model specialists whose instances are launched for exterior verification.
4. DeepSeek-V3.2 is the primary mannequin within the DeepSeek household to combine considering immediately into software use, supporting each considering and non considering software modes and a protocol the place inside reasoning persists throughout software calls and is reset solely on new consumer messages.

Take a look at the Paper and Model weights. Be happy to take a look at our GitHub Page for Tutorials, Codes and Notebooks. Additionally, be happy to comply with us on Twitter and don’t neglect to hitch our 100k+ ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.