DeepSeek
DeepSeek: DeepSeek V3.2
deepseek/deepseek-v3.2
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance.
- Context window: 163,840 tokens
- Input: text
- Output: text
- Pricing: $0.259/M input tokens, $0.42/M output tokens
Community sentiment
Mixed (+2) — 4 positive, 4 negative, 6 neutral across 14 community samples from X and the web, as of 2026-10-04.
Community reactions to DeepSeek V3.2 are mixed: users praise its reasoning and coding value, lower-cost wins on some benchmarks, and recognition at launch. Critics point to inconsistent behavior (model identity/log mismatches), trailing results in blind evaluations, and efficiency/pricing uncertainties after removals from pricing pages.
- “Your Hermes fallback should NOT just be the cheapest model. If your main model fails, the backup still needs to finish the job. 5 Nous Portal options worth looking at: → Gemini 2.5 Flash Lite — general backup → DeepSeek V3.2 — reasoning + coding value → Qwen3.5 Flash — cheap agen” — X
- “Second catch: reasoning models bill you for thinking. On the same coding benchmark, o1 cost 6x more per token than GPT-4o but 27x more per run: $186.50 vs $7.03. It also solved 2.7x more. Nine months later, DeepSeek V3.2 beat o1's score on that test for $0.88. https://t.co/dxoHxz” — X
- “Reran @karpathy's blind-coordinate eval: Claude Sonnet 5.5 hit 118/120; DeepSeek V3.2 and Gemma 4 trailed. https://t.co/nTpFRd5XfG” — X
- “2. The report says that at one million tokens V4-Pro uses 27% of DeepSeek-V3.2’s single-token inference FLOPs and 10% of its KV cache; V4-Flash uses 10% and 7%. It reports strong open-model results across reasoning, coding, browsing, and agent tasks, with the usual caveats of a v” — X
View on OpenRouter.
Model data sourced from OpenRouter.