Coding
Like-for-like- Claude Opus 5
- 75.6
- Supported · #3/183
- DeepSeek V3.2
- 37.2
- Supported · #152/183
- Basis
- BenchAlign lane · 16 vs 2 public rows
- Reading
- Claude Opus 5 leads
Claude Opus 5 vs DeepSeek V3.2
Updated September 4, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
The public evidence has no benchmark result shared by both models, so it does not support a quality verdict. Use the documented cost, context, and runtime rows instead.
0 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.
Share or export
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.
Coding work
Code generation, repair, and software-engineering tasks
Claude Opus 5
Claude Opus 5 leads on the public coding lane, 75.6 to 37.2, with Supported evidence for both models and non-overlapping 90% intervals.
Confidence: stronger
Long documents
Prompts that approach the documented context limit
Claude Opus 5
Claude Opus 5 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
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
DeepSeek V3.2 is not ranked on the public lane for agentic, so no winner is named for agentic.
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.
Confidence: listed-rates
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
73 Claude Opus 5 only
7 DeepSeek V3.2 only
- Shared results
- 0
- Claude Opus 5 only
- 73
- DeepSeek V3.2 only
- 7
- Like-for-like categories
- 1 / 8
1 category rests on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.
Category results, on a stated basis
Each row shows the public-lane category score for both models: the BenchAlign lane for agentic, coding, and knowledge, and the weighted public rows elsewhere, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence or the same weighted set. Directional and not-comparable rows remain visible, but they never receive a winner.
Knowledge
Directional only- Claude Opus 5
- 82.1
- Supported · #3/181
- DeepSeek V3.2
- 49.7
- Estimated · #92/181
- Basis
- BenchAlign lane · 19 vs 0 public rows
- Reading
- Directional only
Agentic
Not comparable- Claude Opus 5
- 77.4
- Supported · #2/151
- DeepSeek V3.2
- Not ranked
- Basis
- BenchAlign lane · 19 vs 3 public rows
- Reading
- Not comparable
Reasoning
Not comparable- Claude Opus 5
- 75.6
- #13/22
- DeepSeek V3.2
- 51.7
- Unranked · 2 rankable rows
- Basis
- Provisional lane · 2 vs 0 weighted rows
- Reading
- Not comparable
Math
Not comparable- Claude Opus 5
- Not ranked
- DeepSeek V3.2
- 40.2
- Unranked · 2 rankable rows
- Basis
- Provisional lane · 0 vs 2 weighted rows
- Reading
- Not comparable
Multilingual
Not comparable- Claude Opus 5
- Not ranked
- DeepSeek V3.2
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Multimodal
Not comparable- Claude Opus 5
- 88.7
- #2/48
- DeepSeek V3.2
- Not ranked
- Basis
- Provisional lane · 1 vs 0 weighted rows
- Reading
- Not comparable
Instruction following
Not comparable- Claude Opus 5
- Not ranked
- DeepSeek V3.2
- 58.0
- #71/120
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
| Category | Claude Opus 5 | DeepSeek V3.2 | Basis | Reading |
|---|---|---|---|---|
| Coding | 75.6Supported · #3/183 | 37.2Supported · #152/183 | Like-for-likeBenchAlign lane · 16 vs 2 public rows | Claude Opus 5 leads |
| Knowledge | 82.1Supported · #3/181 | 49.7Estimated · #92/181 | Directional onlyBenchAlign lane · 19 vs 0 public rows | Directional only |
| Agentic | 77.4Supported · #2/151 | Not ranked | Not comparableBenchAlign lane · 19 vs 3 public rows | Not comparable |
| Reasoning | 75.6#13/22 | 51.7Unranked · 2 rankable rows | Not comparableProvisional lane · 2 vs 0 weighted rows | Not comparable |
| Math | Not ranked | 40.2Unranked · 2 rankable rows | Not comparableProvisional lane · 0 vs 2 weighted rows | Not comparable |
| Multilingual | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Multimodal | 88.7#2/48 | Not ranked | Not comparableProvisional lane · 1 vs 0 weighted rows | Not comparable |
| Instruction following | Not ranked | 58.0#71/120 | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign) differ from the provisional-lane categories. Unranked scores sit on the lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.
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.
A shared-evidence shape is not available.
BenchLM does not draw a radar or infer missing axes when the matched evidence is too sparse.
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 5
- $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 5
- $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 5
- $0.45
- Fits in one request
- DeepSeek V3.2
- $0.0154
- Does not fit in one request
DeepSeek V3.2 does not fit this workload in one request.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationClaude Opus 5DeepSeek V3.2
Context window
Maximum documented context; output-token limits may be lower.
Claude Opus 5
DeepSeek V3.2
128K
API model ID
Claude Opus 5
claude-opus-5
Anthropic model overviewDeepSeek 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 5
$0.5 per 1M cached input tokens
Claude API pricingDeepSeek V3.2
$0.028 per 1M cached input tokens
Documented inputs
Claude Opus 5
text, image
Anthropic model overviewDeepSeek V3.2
Not sourced
Documented outputs
Claude Opus 5
DeepSeek V3.2
Not sourced
Provider availability
Claude Opus 5
Generally Available · Claude API
Anthropic model overviewDeepSeek V3.2
Not sourced
Reasoning profile
Claude Opus 5
Reasoning
DeepSeek V3.2
Non-Reasoning
Weight access
Claude Opus 5
Proprietary
DeepSeek V3.2
Open Weight
License
Claude Opus 5
Proprietary
DeepSeek V3.2
Open Weight
Release date
Claude Opus 5
2026-07-24
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
- The public evidence has no benchmark result shared by both models, so it does not support a quality verdict.
- Workload cost
- Repository review: $0.325 vs $0.01526. Cache-heavy agent loop: $0.45 vs $0.0154.
- Context tradeoff
- Claude Opus 5 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 evidence80 rows
Agentic
Terminal-Bench 3.0
Claude Opus 542.7%SourceDeepSeek V3.2—Not directly comparable
BrowseComp
Claude Opus 590.8%SourceDeepSeek V3.2—Not directly comparable
HLE w/ tools
Claude Opus 564.7%SourceDeepSeek V3.2—Not directly comparable
DeepSearchQA
Claude Opus 595.0%SourceDeepSeek V3.2—Not directly comparable
DRACO
Claude Opus 588.6%SourceDeepSeek V3.2—Not directly comparable
BrowseComp (10-agent, prerelease)
Claude Opus 593.6%SourceDeepSeek V3.2—Not directly comparable
OSWorld 2.0
Claude Opus 570.6%SourceDeepSeek V3.2—Not directly comparable
MCP Atlas
Claude Opus 585.8%SourceDeepSeek V3.2—Not directly comparable
MCP-Atlas claim coverage
Claude Opus 589.1%SourceDeepSeek V3.2—Not directly comparable
LAB all-pass (Anthropic harness)
Claude Opus 523.58%SourceDeepSeek V3.2—Not directly comparable
LAB criterion-pass (Anthropic harness)
Claude Opus 593.74%SourceDeepSeek V3.2—Not directly comparable
LAB all-pass (Harvey held-out)
Claude Opus 511.7%SourceDeepSeek V3.2—Not directly comparable
LAB criterion-pass (Harvey held-out)
Claude Opus 594.1%SourceDeepSeek V3.2—Not directly comparable
Toolathlon-Verified
Claude Opus 580.6%SourceDeepSeek V3.2—Not directly comparable
Toolathlon Verified Pass@3
Claude Opus 587.0%SourceDeepSeek V3.2—Not directly comparable
Toolathlon Verified Pass³
Claude Opus 573.1%SourceDeepSeek V3.2—Not directly comparable
Toolathlon Verified avg. turns
Claude Opus 523.5 turnsSourceDeepSeek V3.2—Not directly comparable
AutomationBench
Claude Opus 526.0%SourceDeepSeek V3.2—Not directly comparable
Terminal-Bench 2.1 (Vals)
Claude Opus 584.6%SourceDeepSeek V3.2—Not directly comparable
Claw-Eval
Claude Opus 5—DeepSeek V3.240.2%SourceNot directly comparable
VITA-Bench
Claude Opus 5—DeepSeek V3.218.5%SourceNot directly comparable
Gert Labs
Claude Opus 5—DeepSeek V3.229.57%SourceNot directly comparable
Coding
Bug Hunt Bench
Claude Opus 527 fixesSourceDeepSeek V3.2—Not directly comparable
SWE-bench Verified
Claude Opus 596%SourceDeepSeek V3.2—Not directly comparable
SWE-bench Pro
Claude Opus 579.2%SourceDeepSeek V3.2—Not directly comparable
SWE Multilingual
Claude Opus 589.5%SourceDeepSeek V3.2—Not directly comparable
SWE Multimodal
Claude Opus 559.4%SourceDeepSeek V3.2—Not directly comparable
deepSwe
Claude Opus 568.8%SourceDeepSeek V3.2—Not directly comparable
FrontierCode 1.1 Main
Claude Opus 553.4%SourceDeepSeek V3.2—Not directly comparable
FrontierCode 1.1 Extended
Claude Opus 563.6%SourceDeepSeek V3.2—Not directly comparable
FrontierSWE v2
Claude Opus 552.0%SourceDeepSeek V3.2—Not directly comparable
ProgramBench (episode 1)
Claude Opus 583.0%SourceDeepSeek V3.2—Not directly comparable
ProgramBench
Claude Opus 593.0%SourceDeepSeek V3.2—Not directly comparable
cursorBench32
Claude Opus 570.0%SourceDeepSeek V3.2—Not directly comparable
VulcanBench v3
Claude Opus 587.0%SourceDeepSeek V3.2—Not directly comparable
VulcanBench CII v1
Claude Opus 596.4%SourceDeepSeek V3.2—Not directly comparable
LiveCodeBench (Vals)
Claude Opus 589.0%SourceDeepSeek V3.2—Not directly comparable
SWE-bench (Vals)
Claude Opus 597.0%SourceDeepSeek V3.2—Not directly comparable
SWE-Rebench
Claude Opus 5—DeepSeek V3.260.9%SourceNot directly comparable
React Native Evals
Claude Opus 5—DeepSeek V3.271.5%SourceNot directly comparable
Reasoning
Knowledge
HLE
Claude Opus 564.7%SourceDeepSeek V3.2—Not directly comparable
HLE w/o tools
Claude Opus 556.3%SourceDeepSeek V3.2—Not directly comparable
HLE-Verified
Claude Opus 554.4%SourceDeepSeek V3.2—Not directly comparable
LABBench2
Claude Opus 584.2%SourceDeepSeek V3.2—Not directly comparable
HealthBench (raw)
Claude Opus 567.1%SourceDeepSeek V3.2—Not directly comparable
HealthBench (length-adjusted)
Claude Opus 557.8%SourceDeepSeek V3.2—Not directly comparable
HealthBench Professional
Claude Opus 559.8%SourceDeepSeek V3.2—Not directly comparable
HealthBench Professional (raw)
Claude Opus 573.4%SourceDeepSeek V3.2—Not directly comparable
BioMysteryBench (human-solvable)
Claude Opus 590.1%SourceDeepSeek V3.2—Not directly comparable
BioMysteryBench (human-difficult)
Claude Opus 549.4%SourceDeepSeek V3.2—Not directly comparable
SpatialBench Verified
Claude Opus 572.5%SourceDeepSeek V3.2—Not directly comparable
SingleCellBench
Claude Opus 560.6%SourceDeepSeek V3.2—Not directly comparable
ProteinGym Hard
Claude Opus 547.7%SourceDeepSeek V3.2—Not directly comparable
Protein Design
Claude Opus 542.5%SourceDeepSeek V3.2—Not directly comparable
Organic chemistry V2
Claude Opus 561.6%SourceDeepSeek V3.2—Not directly comparable
Protocols (troubleshooting)
Claude Opus 561.1%SourceDeepSeek V3.2—Not directly comparable
Protocols (understanding)
Claude Opus 578.4%SourceDeepSeek V3.2—Not directly comparable
GPQA Diamond (Vals)
Claude Opus 593.4%SourceDeepSeek V3.2—Not directly comparable
MMLU-Pro (Vals)
Claude Opus 591.6%SourceDeepSeek V3.2—Not directly comparable
Math
IMO 2026
Claude Opus 542/42SourceDeepSeek V3.2—Not directly comparable
RiemannBench (no tools)
Claude Opus 560.0%SourceDeepSeek V3.2—Not directly comparable
RiemannBench (tools)
Claude Opus 579.0%SourceDeepSeek V3.2—Not directly comparable
ArXivMath Jun. 2026 (no tools)
Claude Opus 590.8%SourceDeepSeek V3.2—Not directly comparable
ArXivMath Jun. 2026 (tools)
Claude Opus 591.3%SourceDeepSeek V3.2—Not directly comparable
FrontierMath v2 (Tiers 1-3)
Claude Opus 5—DeepSeek V3.222.100%SourceNot directly comparable
FrontierMath v2 (Tier 4)
Claude Opus 5—DeepSeek V3.22.100%SourceNot directly comparable
Multilingual
Multimodal
Chartography (no tools)
Claude Opus 529.6%SourceDeepSeek V3.2—Not directly comparable
Chartography (tools)
Claude Opus 583.0%SourceDeepSeek V3.2—Not directly comparable
BenchCAD Vision2Code (no tools)
Claude Opus 50.366SourceDeepSeek V3.2—Not directly comparable
BenchCAD Vision2Code (tools)
Claude Opus 50.821SourceDeepSeek V3.2—Not directly comparable
GDP.pdf (no tools)
Claude Opus 583.4%SourceDeepSeek V3.2—Not directly comparable
GDP.pdf (tools)
Claude Opus 585.5%SourceDeepSeek V3.2—Not directly comparable
OfficeQA
Claude Opus 578.1%SourceDeepSeek V3.2—Not directly comparable
OfficeQA Pro
Claude Opus 566.9%SourceDeepSeek V3.2—Not directly comparable
Frequently asked questions
Which is better, Claude Opus 5 or DeepSeek V3.2?
The public evidence has no benchmark result shared by both models, so it does not support a quality verdict. The page therefore keeps the decision tied to the specific documented workload.
Which is better for coding, Claude Opus 5 or DeepSeek V3.2?
Claude Opus 5 leads the public coding lane, 75.6 to 37.2, with Supported evidence for both models and non-overlapping 90% intervals.
Which is better for agentic tasks, Claude Opus 5 or DeepSeek V3.2?
DeepSeek V3.2 is not ranked on the public lane for agentic tasks, so no winner is named for agentic tasks.
Which costs less, Claude Opus 5 or DeepSeek V3.2?
For the stated presets, chat costs $0.0175 on Claude Opus 5 and $0.00049 on DeepSeek V3.2; repository review costs $0.325 and $0.01526; the cache-heavy agent loop costs $0.45 and $0.0154. DeepSeek V3.2 does not fit this workload in one request.
Which has the larger context window, Claude Opus 5 or DeepSeek V3.2?
Claude Opus 5 has the larger documented context window: 1M, compared with 128K.
Related comparisons
Last updated September 4, 2026
Watch Claude Opus 5 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.