DeepSeekEvidence ready

DeepSeek-V3.2

DeepSeek / DeepSeek-V3.2 · deepseek-ai/DeepSeek-V3.2

Reasoning-first open model built for agents with thinking in tool-use.

Objective score

50

Confidence

92%

Price evidence

Pricing evidence pending

Support this model Compare trend Official source

Other primary tiers in this family

DeepSeek R1 0528 Qwen3 8B DeepSeek R1 Distill Llama 70B DeepSeek R1 Distill Llama 8B DeepSeek R1 Distill Qwen 1.5B DeepSeek R1 Distill Qwen 14B DeepSeek R1 Distill Qwen 32B DeepSeek R1 Distill Qwen 7B DeepSeek V2.5 1210 DeepSeek-R1-0528 DeepSeek-V3-0324DeepSeek-V3.2DeepSeek-V3.2-Speciale

Scenario evidence

Six task-specific rankings

Feature development#17Obj. 500 · 0%Debug and repair#14Obj. 620 · 0%Refactoring#17Obj. 59.60 · 0%Code review#25Obj. 57.40 · 0%Large-repo agent work#25Obj. 38.50 · 0%Frontend / UI#25Obj. 57.20 · 0%

Evidence state

Signal coverage

Benchmarks: 4
Rank points: 7
Usage: 2
Agent: 0
Sources: 12

Objective evidence

Benchmarks and public signals

Swebench: 70
Livecodebench: —
Aider: 70.2
Artificial Analysis: 44.2
Openrouter Usage: 0.6
Kilo Usage: —

Persistent history

Recent ranking points

Sep 15#13 · 46

Sep 15#13 · 46

Sep 15#13 · 46

Sep 14#13 · 46

Sep 14#13 · 46

Sep 14#13 · 46

Usage signals

Overall0.7%

OpenRouter0.6%

Agent and tool signals

Evidence is being collected. Missing data is not treated as zero.

Source trail

Official DeepSeek release says V3.2 is live on app, web, API and open-sourced on Hugging Face.SWE-bench Verified · ok Artificial Analysis Coding Agents · ok Aider polyglot · ok WebDev Arena · ok Kilo leaderboard · ok OpenRouter rankings · ok DeepSeek V3.2 release · ok Step 3.5 Flash model card · ok NVIDIA Nemotron 3 Ultra model card · ok GLM-4.7 model card · ok Kimi K2.6 model card · ok LongCat 2.0 model card · ok