DeepSeek Releases V3.2 and V3.2-Speciale With Thinking in Tool-Use Support

DeepSeek has announced an API update introducing two new model variants, DeepSeek V3.2 and DeepSeek V3.2 Speciale. The announcement, made via the official X account (@deepseek ai) on Friday, July 31, 2026, marks the company's continued push into reasoning based models for AI agent applications. According to the X post, DeepSeek V3.2 follows the same usage pattern as the experimental V3.2 Exp version. Meanwhile, DeepSeek V3.2 Speciale is served through a temporary endpoint with the base URL "https://api.deepseek.com/v3.2 speciale expires on 20251215". This special model is priced the same as V3.2, does not support tool calls, and is only available until December 15, 2025, at 15:59 UTC. Official DeepSeek documentation at api docs.deepseek.com describes both models as "Reasoning first models built for agents," emphasizing their focus on reasoning and agent oriented design.

The release underscores DeepSeek's growing emphasis on developing models capable of complex reasoning and more effective interaction with external tools. A key feature highlighted in the announcement is Thinking in Tool Use support on V3.2. This capability allows the model to reason before or during tool invocation, a critical function for agent scenarios requiring multi step planning. Technical details on this feature are available in DeepSeek's official guide, linked in the announcement. The introduction of V3.2 Speciale with a temporary endpoint reflects DeepSeek's strategy of providing limited access to specific models for a defined period. With pricing identical to V3.2 but lacking tool call support, this variant appears aimed at users who need pure reasoning capabilities without external tool integration.

The usage deadline of December 15, 2025, signals that this model may be a special edition or an experiment before further refinement. From a technical standpoint, V3.2's usage pattern matching V3.2 Exp eases migration for developers already familiar with the experimental version, reducing adoption barriers and speeding integration into existing applications. Third party technical guides, such as those from apidog.com, have already covered how to use both models, including API configuration, authentication, pricing details, advanced reasoning modes, and tool use integration. This indicates that the developer ecosystem has responded to the launch by providing learning resources that accelerate adoption. The release of V3.2 and V3.2 Speciale comes amid intense competition in the large AI model market, where reasoning capabilities and tool integration are key differentiators.

Models that can plan, invoke tools, and execute tasks autonomously are increasingly sought by companies building AI agent solutions for business process automation. For developers and enterprises using the DeepSeek API, this update offers additional options to balance cost and capability. V3.2 with Thinking in Tool Use provides richer functionality for agent applications, while V3.2 Speciale offers an alternative at the same price but without tool calls, suitable for use cases focused on pure reasoning. The limited availability of V3.2 Speciale until December 15, 2025, also creates urgency for developers who wish to evaluate the model before its endpoint is shut down. After that date, users requiring similar capabilities must switch to V3.2 or wait for a subsequent variant that DeepSeek may release.

Official DeepSeek documentation describes V3.2 as part of the evolution of models designed for agents, reaffirming the company's commitment to developing more autonomous and reasoning capable AI. With Thinking in Tool Use support, DeepSeek aims to meet market demand for models that not only generate text but also intelligently plan and execute actions through external tools. This API update also reflects a broader industry trend where the line between language models and executing agents is blurring. Major AI companies are racing to deliver models that can serve as the brain of autonomous systems, whether for customer service, data analysis, or complex workflow automation.

With identical pricing for V3.2 and V3.2 Speciale, DeepSeek appears not to differentiate pricing based on tool call capability, but rather on availability and validity period. This strategy allows users to test the model at the same cost while providing DeepSeek with valuable usage data for future development. Looking ahead, developers interested in using V3.2 Speciale must note the December 15, 2025 deadline and plan migration to V3.2 once the temporary endpoint is closed. Meanwhile, V3.2 with Thinking in Tool Use support emerges as a more stable long term option for production applications. This launch adds to DeepSeek's growing list of innovations in open AI models and APIs, strengthening its position as a key player in the global AI ecosystem.

With a focus on reasoning and agent capabilities, DeepSeek targets a market segment that requires intelligent models for tasks involving multi level planning and execution.