DeepSeek-V3.2: benchmark scores, price and ranking

DeepSeek-V3.2 from DeepSeek ranks #406 of 743 LLMs on the LLMs Tiger aggregate benchmark score (47.7), based on 4 benchmark results captured up to 2026-10-13.

Organisation: DeepSeek
Type: Open-weight / local model
Open weights: Yes
Released: 2025-12-01 (source: curated)
Context window: 164,000 tokens
Cheapest API price: $0.259 in / $0.800 out per million tokens (via openrouter, observed 2026-10-13T20:47Z)
Aggregate rank: #406 of 743 (score 47.7)

Relative to other models, DeepSeek-V3.2 is strongest on Epoch ECI (#23 of 204) and weakest on SWE-bench Multilingual (#11 of 11).

Across the 4 benchmarks where at least ten models are scored, DeepSeek-V3.2 is in the top 10% on 0 and in the top half on 2.

Its cheapest listed input price, $0.259 per million tokens, is lower than that of 63% of the 427 models with a listed price.

DeepSeek-V3.2 benchmark results

DeepSeek-V3.2 benchmark scores, ranked against all models measured on each benchmark
Benchmark	Score	Rank	Source	Captured
Epoch ECI	146	#23 of 204	Epoch AI Benchmarking Hub	2026-10-13
SWE-bench Verified	70.0%	#20 of 73	SWE-bench Verified leaderboard	2026-10-13
ARC-AGI-2	4.0%	#63 of 87	ARC Prize official leaderboard	2026-10-13
SWE-bench Multilingual	59.0%	#11 of 11	SWE-bench Multilingual leaderboard	2026-10-13

About DeepSeek-V3.2

DeepSeek-V3.2 is an open-weight model that can run on your own hardware from DeepSeek. It was released on 2025-12-01, according to the developer's announcement. It accepts up to 164,000 tokens of input at once. The cheapest API listing we track charges $0.259 per million input tokens and $0.800 per million output tokens, through openrouter. We hold 4 benchmark results for it across 3 areas, and none is in the top tenth of its board.

Where DeepSeek-V3.2 stands

On our aggregate score, DeepSeek-V3.2 is #406 of 743 ranked models, around the middle of the table. Its score is 47.7; the scale is centred on 50, so it sits 2.3 points below the typical model. That beats 45% of the ranked field. Among the 33 ranked DeepSeek models it is #22. The next one up is DeepSeek-V3.2 (Thinking; Fireworks) at 47.9. The next one down is Deepseek R1 (05/28) at 47.5.

DeepSeek-V3.2, benchmark by benchmark

Epoch ECI puts DeepSeek-V3.2 at 146, #23 of 204. That beats 89% of the models tested, but 10 points short of Qwen 3.8 Max (156). The results either side are Grok 4 (146) and nemotron-3-ultra (146). It covers Epoch Capabilities Index: Epoch's own aggregate of ~40 benchmarks. The board median is 127, so DeepSeek-V3.2 is 19 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.

On SWE-bench Verified, DeepSeek-V3.2 reaches 70.0%. It ranks #20 of 73, ahead of 73% of the field, and 13.5 points behind Claude Opus 4.7 (max) at 83.5%. The results either side are Claude 4.5 Sonnet (70.6%) and Gemini 3 Pro (69.6%). It covers real GitHub issues resolved. The board median is 56.4%, so DeepSeek-V3.2 is 13.6 points above it. Source: SWE-bench Verified leaderboard, captured 2026-10-13.

DeepSeek-V3.2 lands below the middle on ARC-AGI-2: 4.0%, #63 of 87, with 71% of the field scoring higher. GPT-6 Astra (max) leads at 95.0%; the closest result above is Claude Haiku 4.5 (Thinking 32K) at 4.0%. It covers abstract reasoning, ARC Prize public eval set. The board median is 22.8%, so DeepSeek-V3.2 is 18.8 points below it. Source: ARC Prize official leaderboard, captured 2026-10-13.

SWE-bench Multilingual is a weak spot. DeepSeek-V3.2 scores 59.0%, #11 of 11, behind 91% of the models measured. The leader, Gemini 3 Flash, reaches 72.7%. It covers real GitHub issues resolved. The board median is 68.3%, so DeepSeek-V3.2 is 9.3 points below it. Source: SWE-bench Multilingual leaderboard, captured 2026-10-13.

Strengths and weaknesses

General knowledge and chat: uneven, ahead of 44% of the field on average across 2 benchmarks. Its best board here is Epoch ECI (#23 of 204) and its weakest is SWE-bench Multilingual (#11 of 11).

Price and value

The median input price across the 427 priced models is $0.500 per million tokens, so DeepSeek-V3.2 costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on DeepSeek-V3.1-Terminus ($0.270) and MiniMax-M2.5 ($0.270).

Release history

The DeepSeek model released before it was deepseek-chat on 2025-12-01, which scores 46.5. The next one was DeepSeek-V3.2-Speciale on 2025-12-01, at 50.4.

Reading these numbers

The 4 scores on this page come from 4 sources: ARC Prize official leaderboard (1), Epoch AI Benchmarking Hub (1), SWE-bench Multilingual leaderboard (1), SWE-bench Verified leaderboard (1). 2 of them come from an official leaderboard and 2 from datasets or third-party runs. The leaderboard also lists 1 other reasoning-effort setting of this model, which redirect here. This page shows the highest. With 4 results the aggregate rank can move a lot when one new score arrives.

DeepSeek-V3.2 is also listed at 1 other reasoning-effort setting; this page shows the highest-effort row, which is the one the leaderboard keeps by default.

Models ranked nearby

#403 Tiny Recursion Model (TRM) (47.8)
#404 DeepSeek-V2-Chat (47.8)
#405 Muse Code + Muse Spark 1.3 (47.8)
#407 Grok 4 (Fast Reasoning) (47.7)
#408 Mistral-Small-24B-Base-2501 (47.7)
#409 Gemini 2.5 Pro (Thinking 32K) (47.7)