Math
Like-for-like- Claude Opus 4.6
- 36.3
- DeepSeek V3.2
- 17.1
- Weighted basis
- 2 vs 2 rows
- Reading
- Claude Opus 4.6 leads
Claude Opus 4.6 vs DeepSeek V3.2
Updated September 3, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
Claude Opus 4.6 has the higher public score estimate, 67.98 versus 54.75, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
6 results are shared. Category rows based on different benchmark sets are marked directional and do not name a winner.
Which one for your work
Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.
Long documents
Prompts that approach the documented context limit
Claude Opus 4.6
Claude Opus 4.6 has the larger documented context window.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
DeepSeek V3.2
DeepSeek V3.2 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Show secondary and unsupported calls
Repository review cost
50K fresh input + 3K output tokens
DeepSeek V3.2
DeepSeek V3.2 has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
No shared weighted benchmark basis supports a winner.
Confidence: limited
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Not enough matched evidence
The page does not recommend a cost winner because at least one model cannot fit the stated workload in one request. DeepSeek V3.2 does not fit this workload in one request. Claude Opus 4.6 has no published cached-input rate, so cached tokens use its listed input rate.
Confidence: rate-fallback
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
27 Claude Opus 4.6 only
6 Shared
1 DeepSeek V3.2 only
- Shared results
- 6
- Claude Opus 4.6 only
- 27
- DeepSeek V3.2 only
- 1
- Like-for-like categories
- 1 / 8
1 category uses different evidence sets. Those rows remain visible for coverage context but do not name a winner.
Category results, on a stated basis
Each row states whether both averages use the same weighted benchmark set. Directional and not-comparable rows remain visible, but they never receive a winner in this template.
Coding
Directional only- Claude Opus 4.6
- 68.1
- DeepSeek V3.2
- 60.9
- Weighted basis
- 3 vs 1 rows
- Reading
- Directional only
Agentic
Not comparable- Claude Opus 4.6
- 73.0
- DeepSeek V3.2
- Not measured
- Weighted basis
- 3 vs 0 rows
- Reading
- Not comparable
Reasoning
Not comparable- Claude Opus 4.6
- Not measured
- DeepSeek V3.2
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Knowledge
Not comparable- Claude Opus 4.6
- 69.1
- DeepSeek V3.2
- Not measured
- Weighted basis
- 4 vs 0 rows
- Reading
- Not comparable
Multilingual
Not comparable- Claude Opus 4.6
- Not measured
- DeepSeek V3.2
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Multimodal
Not comparable- Claude Opus 4.6
- 77.3
- DeepSeek V3.2
- Not measured
- Weighted basis
- 1 vs 0 rows
- Reading
- Not comparable
Instruction following
Not comparable- Claude Opus 4.6
- Not measured
- DeepSeek V3.2
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
| Category | Claude Opus 4.6 | DeepSeek V3.2 | Weighted basis | Reading |
|---|---|---|---|---|
| Math | 36.3 | 17.1 | Like-for-like2 vs 2 rows | Claude Opus 4.6 leads |
| Coding | 68.1 | 60.9 | Directional only3 vs 1 rows | Directional only |
| Agentic | 73.0 | Not measured | Not comparable3 vs 0 rows | Not comparable |
| Reasoning | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Knowledge | 69.1 | Not measured | Not comparable4 vs 0 rows | Not comparable |
| Multilingual | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Multimodal | 77.3 | Not measured | Not comparable1 vs 0 rows | Not comparable |
| Instruction following | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
Shape of the matched evidence
Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.
Too few matched category axes support a radar. The ruled list below shows only shared benchmark results; positions use each benchmark’s normalized display scale when available.
FrontierMath v2 (Tier 4)
Math
FrontierMath v2 (Tiers 1-3)
Math
SWE-Rebench
Coding
Claude Opus 4.6: 65.3%DeepSeek V3.2: 60.9%Normalized gap 4.4Claude Opus 4.6 source DeepSeek V3.2 source
What each workload costs
Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.
Chat turn
1K fresh input + 500 output tokens
- Claude Opus 4.6
- $0.0175
- Fits in one request
- DeepSeek V3.2
- $0.00049
- Fits in one request
DeepSeek V3.2 has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- Claude Opus 4.6
- $0.325
- Fits in one request
- DeepSeek V3.2
- $0.01526
- Fits in one request
DeepSeek V3.2 has the lower modeled cost
Costs use the listed standard API rates.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- Claude Opus 4.6
- $1.35
- Fits in one request
- Cached input priced at the published list-input rate
- DeepSeek V3.2
- $0.0154
- Does not fit in one request
DeepSeek V3.2 does not fit this workload in one request. Claude Opus 4.6 has no published cached-input rate, so cached tokens use its listed input rate.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationClaude Opus 4.6DeepSeek V3.2
Context window
Maximum documented context; output-token limits may be lower.
Claude Opus 4.6
1M
DeepSeek V3.2
128K
API model ID
Claude Opus 4.6
Not sourced
DeepSeek V3.2
Not sourced
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Claude Opus 4.6
Not published
DeepSeek V3.2
$0.028 per 1M cached input tokens
Documented inputs
Claude Opus 4.6
Not sourced
DeepSeek V3.2
Not sourced
Documented outputs
Claude Opus 4.6
Not sourced
DeepSeek V3.2
Not sourced
Provider availability
Claude Opus 4.6
Not sourced
DeepSeek V3.2
Not sourced
Reasoning profile
Claude Opus 4.6
Non-Reasoning
DeepSeek V3.2
Non-Reasoning
Weight access
Claude Opus 4.6
Proprietary
DeepSeek V3.2
Open Weight
License
Claude Opus 4.6
Proprietary
DeepSeek V3.2
Open Weight
Release date
Claude Opus 4.6
2026-02-01
DeepSeek V3.2
2025-12-01
If you already use one of these models
- Deployment change
- The models list different providers, so authentication, endpoint behavior, limits, and feature support may change.
- Quality signal
- Claude Opus 4.6 has the higher public score estimate, 67.98 versus 54.75, but the 90% score intervals overlap.
- Workload cost
- Repository review: $0.325 vs $0.01526. Cache-heavy agent loop: $1.35 vs $0.0154.
- Context tradeoff
- Claude Opus 4.6 has the larger documented window (1M).
Run the same representative tasks against both endpoints before changing production traffic.
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence34 rows
Agentic
Terminal-Bench 2.0
Claude Opus 4.665.4%SourceDeepSeek V3.2—Not directly comparable
BrowseComp
Claude Opus 4.683.7%SourceDeepSeek V3.2—Not directly comparable
OSWorld-Verified
Claude Opus 4.672.7%SourceDeepSeek V3.2—Not directly comparable
Claw-Eval
Shared sourceClaude Opus 4.670.4%DeepSeek V3.240.2%Claude Opus 4.6 leads this result
DeepSearchQA
Claude Opus 4.673.7%SourceDeepSeek V3.2—Not directly comparable
CyberGym
Claude Opus 4.666.6%SourceDeepSeek V3.2—Not directly comparable
Gert Labs
Shared sourceClaude Opus 4.661.85%DeepSeek V3.229.57%Claude Opus 4.6 leads this result
ResearchClawBench
Claude Opus 4.619.9%SourceDeepSeek V3.2—Not directly comparable
JobBench
Claude Opus 4.636.7%SourceDeepSeek V3.2—Not directly comparable
VITA-Bench
Claude Opus 4.6—DeepSeek V3.218.5%SourceNot directly comparable
Coding
SWE-bench Verified
Claude Opus 4.680.8%SourceDeepSeek V3.2—Not directly comparable
SWE-bench Verified*
Claude Opus 4.675.6%SourceDeepSeek V3.2—Not directly comparable
LiveCodeBench Pro
Claude Opus 4.670.7%SourceDeepSeek V3.2—Not directly comparable
SWE-bench Pro
Claude Opus 4.653.4%SourceDeepSeek V3.2—Not directly comparable
SWE-Rebench
Claude Opus 4.665.3%SourceDeepSeek V3.260.9%SourceClaude Opus 4.6 leads this result
React Native Evals
Shared sourceClaude Opus 4.684.1%DeepSeek V3.271.5%Claude Opus 4.6 leads this result
Vibe Code Bench
Claude Opus 4.657.57%SourceDeepSeek V3.2—Not directly comparable
FrontierCode 1.1 Main
Claude Opus 4.626.9%SourceDeepSeek V3.2—Not directly comparable
Knowledge
GPQA
Claude Opus 4.691.3%SourceDeepSeek V3.2—Not directly comparable
GPQA-D
Claude Opus 4.689.2%SourceDeepSeek V3.2—Not directly comparable
SuperGPQA
Claude Opus 4.695%SourceDeepSeek V3.2—Not directly comparable
MMLU-Pro
Claude Opus 4.682%SourceDeepSeek V3.2—Not directly comparable
MMLU-Pro (Arcee)
Claude Opus 4.689.1%SourceDeepSeek V3.2—Not directly comparable
HLE
Claude Opus 4.653%SourceDeepSeek V3.2—Not directly comparable
HLE w/o tools
Claude Opus 4.640%SourceDeepSeek V3.2—Not directly comparable
HealthBench Hard
Claude Opus 4.614.8%SourceDeepSeek V3.2—Not directly comparable
MedXpertQA (Text)
Claude Opus 4.652.1%SourceDeepSeek V3.2—Not directly comparable
Math
AIME25 (Arcee)
Claude Opus 4.699.8%SourceDeepSeek V3.2—Not directly comparable
FrontierMath v2 (Tiers 1-3)
Shared sourceClaude Opus 4.640.700%DeepSeek V3.222.100%Claude Opus 4.6 leads this result
FrontierMath v2 (Tier 4)
Shared sourceClaude Opus 4.622.900%DeepSeek V3.22.100%Claude Opus 4.6 leads this result
Multimodal
Frequently asked questions
Which is better, Claude Opus 4.6 or DeepSeek V3.2?
Claude Opus 4.6 has the higher public score estimate, 67.98 versus 54.75, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
Which is better for coding, Claude Opus 4.6 or DeepSeek V3.2?
The current coding averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which is better for agentic tasks, Claude Opus 4.6 or DeepSeek V3.2?
The published evidence does not provide a shared weighted agentic tasks basis for both models, so BenchLM does not name a agentic tasks winner.
Which costs less, Claude Opus 4.6 or DeepSeek V3.2?
For the stated presets, chat costs $0.0175 on Claude Opus 4.6 and $0.00049 on DeepSeek V3.2; repository review costs $0.325 and $0.01526; the cache-heavy agent loop costs $1.35 and $0.0154. DeepSeek V3.2 does not fit this workload in one request. Claude Opus 4.6 has no published cached-input rate, so cached tokens use its listed input rate.
Which has the larger context window, Claude Opus 4.6 or DeepSeek V3.2?
Claude Opus 4.6 has the larger documented context window: 1M, compared with 128K.
Related comparisons
Last updated September 3, 2026
Watch Claude Opus 4.6 vs DeepSeek V3.2
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.