Model comparison

Claude Opus 4.8 vs DeepSeek-V3.2-Exp

Claude Opus 4.8 is the stronger model overall, scoring 60.7 to 44.9 on the Noometry Index. DeepSeek-V3.2-Exp costs 34× less per token, which makes it the better buy when Claude Opus 4.8's lead doesn't matter for your workload.

Last verified October 10, 2026. 37 shared benchmarks.

Claude Opus 4.8 Anthropic

60.7

Rank #9 Confirmed

DeepSeek-V3.2-Exp

DeepSeek

44.9

Rank #68 Confirmed

Summary

They share 37 benchmarks with published results for both. Claude Opus 4.8 scores higher in 8 categories and DeepSeek-V3.2-Exp in 1 category; 9 gaps are clear of the uncertainty.
The widest gap is in reasoning, where Claude Opus 4.8 leads 64.6 to 22.0.
The biggest single-benchmark swing is ARC-AGI-2: 72.1% for Claude Opus 4.8 and 4% for DeepSeek-V3.2-Exp.
DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $5 / $25 for Claude Opus 4.8.
Claude Opus 4.8 accepts more context: 1M tokens versus 164K.
DeepSeek-V3.2-Exp has downloadable open weights; the other is API-only.

Side by side

Claude Opus 4.8 and DeepSeek-V3.2-Exp specifications
	Claude Opus 4.8	DeepSeek-V3.2-Exp
Provider	Anthropic	DeepSeek
Noometry Index	60.7	44.9
Released	2026-05-28	2025-09-29
Weights	Proprietary	Open
Context window	1M	164K
Max output	128K	66K
Input $ / M tokens	$5	$0.26
Output $ / M tokens	$25	$0.38
Results tracked	65	48

Sponsored placements are available on pages like this one. Advertise on Noometry

Category by category

Coding Claude Opus 4.8 leads

Claude Opus 4.8: 60.2 (#9), DeepSeek-V3.2-Exp: 46.4 (#61)

Coding benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
LMArena WebDev	1556	1362
SciCode	53.5%	38.9%
WeirdML	82.9%	39.5%
LMArena Coding	1491	1454
DeepSWE	59%	—
FrontierCode	46.5%	—
SWE-bench Verified (bash only)	—	70%
Aider Polyglot	—	74.2%
SWE-bench Multilingual	—	59%
GSO	47.1%	—
ALE-Bench	1,564	—

Agentic & Tool Use Claude Opus 4.8 leads

Claude Opus 4.8: 47.6 (#10), DeepSeek-V3.2-Exp: 37.9 (#31)

Agentic & Tool Use benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
Vending-Bench 2	5,787	1,034
Terminal-Bench	—	39.6%
APEX-Agents	48.9%	—
Berkeley Function Calling Leaderboard	—	56.7%
OSWorld 2.0	20.6%	—
Remote Labor Index	8.3%	—
TheAgentCompany	—	42.9%
τ²-bench Banking	39.7%	—
DeepResearch Bench	50.2%	—
PostTrainBench	33.8%	—
GBAEval	70.9%	—
GDP.pdf	24%	—
LMArena Search	1204	—

Reasoning Claude Opus 4.8 leads

Claude Opus 4.8: 64.6 (#14), DeepSeek-V3.2-Exp: 22.0 (#185)

Reasoning benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
ARC-AGI-2	72.1%	4%
Kagi LLM Benchmark	88.8%	52.2%
NYT Connections (extended)	91.1%	36.7%
ARC-AGI-1	92.5%	57%
CritPt	20.9%	2.9%
Chess Puzzles	34%	14%
LMArena Hard Prompts	1481	1433
DTBench	94.9%	87.7%
LMCA	57.5%	29.1%
Epoch Capabilities Index	158.3	146.28
SimpleBench	64.8%	—
EnigmaEval	23.5%	—
Thematic Generalization	—	65%
EBR-Bench	28.6%	—
Mystery Game Puzzles	36%	—
Surface Evolver Bench	87.5%	—
Bench to the Future 3	0.14	—
ForecastBench	59.9	—

Math Claude Opus 4.8 leads

Claude Opus 4.8: 78.2 (#11), DeepSeek-V3.2-Exp: 40.8 (#75)

Math benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
MathArena Final-Answer Competitions	91.8%	57.7%
OTIS Mock AIME 2024-2025	98.3%	87.8%
ProofBench	69%	8%
LMArena Math	1487	1435
FrontierMath (Feb 2025 set)	47.2%	22.1%
FrontierMath Tier 4 (v1)	31.3%	2.1%
FrontierMath (Tiers 1-3)	80%	—
FrontierMath Tier 4	56.1%	—

Knowledge Claude Opus 4.8 leads

Claude Opus 4.8: 61.3 (#26), DeepSeek-V3.2-Exp: 51.6 (#61)

Knowledge benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
GPQA Diamond	91%	83.4%
LMArena Expert	1501	1436
SimpleQA Verified	53%	—
Vectara Hallucination Rate	—	5.3%

Multimodal Not comparable

Claude Opus 4.8: 43.2 (#20), DeepSeek-V3.2-Exp: —

Multimodal benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
LMArena Vision	1295	—
Blueprint-Bench 2	14.5%	—
Furniture Assembly	42.5%	—
LMArena Document	1475	—

Multilingual Claude Opus 4.8 leads

Claude Opus 4.8: 55.4 (#31), DeepSeek-V3.2-Exp: 52.4 (#85)

Multilingual benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
LMArena Non-English	1450	1409
LMArena Chinese	1509	1461
LMArena French	1482	1435
LMArena German	1470	1440
LMArena Japanese	1440	1377
LMArena Korean	1432	1372
LMArena Russian	1476	1423
LMArena Spanish	1470	1441

Instruction Following Claude Opus 4.8 leads

Claude Opus 4.8: 77.6 (#22), DeepSeek-V3.2-Exp: 74.7 (#89)

Instruction Following benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
LMArena Instruction Following	1476	1413

Long Context DeepSeek-V3.2-Exp leads

Claude Opus 4.8: 45.4 (#37), DeepSeek-V3.2-Exp: 47.6 (#15)

Long Context benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
LMArena Longer Query	1482	1428
Fiction.LiveBench	—	83.3%
CL-bench	—	13.2%
CL-bench Life	—	9.5%

Writing & Preference Claude Opus 4.8 leads

Claude Opus 4.8: 72.1 (#17), DeepSeek-V3.2-Exp: 62.5 (#73)

Writing & Preference benchmarks
Benchmark	Claude Opus 4.8	DeepSeek-V3.2-Exp
LMArena Text	1460	1425
LMArena Creative Writing	1453	1403
EQ-Bench Creative Writing	1840	1515
LMArena Multi-Turn	1476	1427
EQ-Bench 4	1281	—

Frequently asked questions

Is Claude Opus 4.8 better than DeepSeek-V3.2-Exp?

Which is cheaper, Claude Opus 4.8 or DeepSeek-V3.2-Exp?

DeepSeek-V3.2-Exp is cheaper. It lists at $0.26 per million input tokens and $0.38 per million output tokens; Claude Opus 4.8 lists at $5 and $25.

Is Claude Opus 4.8 or DeepSeek-V3.2-Exp better for coding?

Claude Opus 4.8 scores higher on coding benchmarks: 60.2 versus 46.4 in the Noometry coding category.

Which has the bigger context window?

Claude Opus 4.8 does, with 1M tokens against 164K.

How many benchmarks do Claude Opus 4.8 and DeepSeek-V3.2-Exp share?

37 benchmarks have published results for both models. Claude Opus 4.8 has 65 scored results on Noometry and DeepSeek-V3.2-Exp has 48.