DeepSeek announces DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, reasoning-first models for agents. V3.2 is the successor to V3.2-Exp and is available on App, Web, and API. V3.2-Speciale is API-only and focuses on maximum reasoning capabilities, achieving gold-level results in IMO, CMO, ICPC World Finals, and IOI 2025. The models introduce thinking in tool-use and a new agent training data synthesis method.

New model

How it measures

Artificial Analysis · Intelligence Index · Reasoning

21.5219thof 697 models

Coding index44.2 · 129th

Math index92 · 17th

Median 12.5Top 57.6

Benchmarks, as Artificial Analysis ran them

84%GPQA Diamond
25%Humanity's Last Exam
47%Terminal-Bench 2.1
36%Terminal-Bench Hard
86%LiveCodeBench
92%AIME 2025
91%τ²-Bench
86%MMLU-Pro
61%IFBench
73%Long context reasoning

Per 1M tokens: $0.28 in · $0.42 out
Context: 164K tokens

LMArena · Text125thof 417· 34th of 227 open models

1Gemini 4 Argon (High)
2Muse Wonder (Max)
3Claude Opus 5.5 (High)
4Claude Fable 5 (High)

DeepSeek-V3.2 · 1,425 (1,421–1,428) · 48,095 votes

Against list priceRating for the price

DeepSeek-V3.2 · 1,425 · $0.32 per 1M tokens ($0.28 in, $0.42 out) · 164K context

LMArena, CC BY 4.0, 20 October 2026 · prices OpenRouter, list price blended 3:1, read 22 October 2026 · Artificial Analysis, read 22 October 2026

About the model

Lab: DeepSeek
Variant: DeepSeek V3.2 Exp
Price: Same pricing as V3.2
OpenRouter: Listed 1 December 2025deepseek/deepseek-v3.2

Ask your AI

Your brief, straight from the record

Connect once, then ask Claude or ChatGPT what changed for the tools you use. Every item links to its page on forck.live.

+ any MCP app

Connect your AI →