Multimodal
Like-for-like- Claude Opus 5
- 66.9
- Hy4 preview
- 66.2
- Weighted basis
- 1 vs 1 rows
- Reading
- Claude Opus 5 leads
Claude Opus 5 vs Hy4 preview
Updated August 28, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
Decision reading
Claude Opus 5 has the higher public score estimate, 83.06 versus 79.16, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
12 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.
No workload recommendation clears the current evidence threshold.
Use the matched evidence, workload costs, and sourced specifications below instead of treating a point score as a universal answer.
Show secondary and unsupported calls
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
Long documents
Prompts that approach the documented context limit
No clear pick
The documented context windows are equal.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
Repository review cost
50K fresh input + 3K output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
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.
52 Claude Opus 5 only
12 Shared
16 Hy4 preview only
- Shared results
- 12
- Claude Opus 5 only
- 52
- Hy4 preview only
- 16
- Like-for-like categories
- 1 / 8
2 categories use 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 5
- 89.5
- Hy4 preview
- 65.7
- Weighted basis
- 2 vs 1 rows
- Reading
- Directional only
Knowledge
Directional only- Claude Opus 5
- 64.7
- Hy4 preview
- 60.4
- Weighted basis
- 1 vs 2 rows
- Reading
- Directional only
Agentic
Not comparable- Claude Opus 5
- 90.8
- Hy4 preview
- Not measured
- Weighted basis
- 1 vs 0 rows
- Reading
- Not comparable
Reasoning
Not comparable- Claude Opus 5
- 90.4
- Hy4 preview
- Not measured
- Weighted basis
- 1 vs 0 rows
- Reading
- Not comparable
Math
Not comparable- Claude Opus 5
- Not measured
- Hy4 preview
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Multilingual
Not comparable- Claude Opus 5
- Not measured
- Hy4 preview
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Instruction following
Not comparable- Claude Opus 5
- Not measured
- Hy4 preview
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
| Category | Claude Opus 5 | Hy4 preview | Weighted basis | Reading |
|---|---|---|---|---|
| Multimodal | 66.9 | 66.2 | Like-for-like1 vs 1 rows | Claude Opus 5 leads |
| Coding | 89.5 | 65.7 | Directional only2 vs 1 rows | Directional only |
| Knowledge | 64.7 | 60.4 | Directional only1 vs 2 rows | Directional only |
| Agentic | 90.8 | Not measured | Not comparable1 vs 0 rows | Not comparable |
| Reasoning | 90.4 | Not measured | Not comparable1 vs 0 rows | Not comparable |
| Math | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Multilingual | Not measured | Not measured | Not comparable0 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.
SWE-bench Pro
Coding
HLE
Knowledge
OfficeQA Pro
Multimodal
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
- Hy4 preview
- Self-hosted; infrastructure cost varies
- Fits in one request
Hy4 preview has no comparable published API token rate.
Repository review
50K fresh input + 3K output tokens
- Claude Opus 5
- $0.325
- Fits in one request
- Hy4 preview
- Self-hosted; infrastructure cost varies
- Fits in one request
Hy4 preview has no comparable published API token rate.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- Claude Opus 5
- $0.45
- Fits in one request
- Hy4 preview
- Self-hosted; infrastructure cost varies
- Fits in one request
- Cached-input rate unavailable
Hy4 preview has no comparable published API token rate.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationClaude Opus 5Hy4 preview
Context window
Maximum documented context; output-token limits may be lower.
Claude Opus 5
Hy4 preview
API model ID
Claude Opus 5
claude-opus-5
Anthropic model overviewHy4 preview
tencent/Hy4-preview
Tencent Hy4 preview model cardCached-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 pricingHy4 preview
No comparable hosted API rate
Tencent Hy4 preview model cardDocumented inputs
Claude Opus 5
text, image
Anthropic model overviewHy4 preview
Not sourced
Documented outputs
Claude Opus 5
Hy4 preview
Not sourced
Provider availability
Claude Opus 5
Generally Available · Claude API
Anthropic model overviewHy4 preview
Not sourced
Reasoning profile
Claude Opus 5
Reasoning
Hy4 preview
Reasoning
Weight access
Claude Opus 5
Proprietary
Hy4 preview
Open Weight
License
Claude Opus 5
Proprietary
Hy4 preview
Open Weight
Release date
Claude Opus 5
2026-07-24
Hy4 preview
2026-08-28
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 5 has the higher public score estimate, 83.06 versus 79.16, but the 90% score intervals overlap.
- Workload cost
- A complete comparable API-rate estimate is not available for both models.
- Context tradeoff
- Both models list 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%SourceHy4 preview—Not directly comparable
BrowseComp
Claude Opus 590.8%SourceHy4 preview—Not directly comparable
HLE w/ tools
Claude Opus 564.7%SourceHy4 preview55.4%SourceClaude Opus 5 leads this result
DeepSearchQA
Claude Opus 595.0%SourceHy4 preview—Not directly comparable
DRACO
Claude Opus 588.6%SourceHy4 preview77.2%SourceClaude Opus 5 leads this result
BrowseComp (10-agent, prerelease)
Claude Opus 593.6%SourceHy4 preview—Not directly comparable
OSWorld 2.0
Claude Opus 570.6%SourceHy4 preview—Not directly comparable
MCP Atlas
Claude Opus 585.8%SourceHy4 preview83.7%SourceClaude Opus 5 leads this result
MCP-Atlas claim coverage
Claude Opus 589.1%SourceHy4 preview—Not directly comparable
LAB all-pass (Anthropic harness)
Claude Opus 523.58%SourceHy4 preview—Not directly comparable
LAB criterion-pass (Anthropic harness)
Claude Opus 593.74%SourceHy4 preview—Not directly comparable
LAB all-pass (Harvey held-out)
Claude Opus 511.7%SourceHy4 preview—Not directly comparable
LAB criterion-pass (Harvey held-out)
Claude Opus 594.1%SourceHy4 preview—Not directly comparable
Toolathlon-Verified
Claude Opus 580.6%SourceHy4 preview74.1%SourceClaude Opus 5 leads this result
Toolathlon Verified Pass@3
Claude Opus 587.0%SourceHy4 preview—Not directly comparable
Toolathlon Verified Pass³
Claude Opus 573.1%SourceHy4 preview—Not directly comparable
Toolathlon Verified avg. turns
Claude Opus 523.5 turnsSourceHy4 preview—Not directly comparable
AutomationBench
Claude Opus 526.0%SourceHy4 preview32.1%SourceHy4 preview leads this result
Terminal-Bench 2.1
Claude Opus 5—Hy4 preview85.4%SourceNot directly comparable
CyberGym
Claude Opus 5—Hy4 preview78.4%SourceNot directly comparable
WideResearch
Claude Opus 5—Hy4 preview83.9%SourceNot directly comparable
APEX-Agents
Claude Opus 5—Hy4 preview37.1%SourceNot directly comparable
skillsBench
Claude Opus 5—Hy4 preview62.9%SourceNot directly comparable
JobBench
Claude Opus 5—Hy4 preview61.7%SourceNot directly comparable
Agents' Last Exam
Claude Opus 5—Hy4 preview22.8%SourceNot directly comparable
BankerToolBench
Claude Opus 5—Hy4 preview78.6%SourceNot directly comparable
Coding
SWE-bench Verified
Claude Opus 596%SourceHy4 preview—Not directly comparable
SWE-bench Pro
Claude Opus 579.2%SourceHy4 preview65.7%SourceClaude Opus 5 leads this result
SWE Multilingual
Claude Opus 589.5%SourceHy4 preview82.9%SourceClaude Opus 5 leads this result
SWE Multimodal
Claude Opus 559.4%SourceHy4 preview—Not directly comparable
deepSwe
Claude Opus 568.8%SourceHy4 preview64.3%SourceClaude Opus 5 leads this result
FrontierCode 1.1 Main
Claude Opus 553.4%SourceHy4 preview—Not directly comparable
FrontierCode 1.1 Extended
Claude Opus 563.6%SourceHy4 preview—Not directly comparable
ProgramBench (episode 1)
Claude Opus 583.0%SourceHy4 preview—Not directly comparable
ProgramBench
Claude Opus 593.0%SourceHy4 preview17.5%SourceClaude Opus 5 leads this result
cursorBench32
Claude Opus 570.0%SourceHy4 preview—Not directly comparable
VulcanBench v3
Claude Opus 587.0%SourceHy4 preview—Not directly comparable
VulcanBench CII v1
Claude Opus 596.4%SourceHy4 preview—Not directly comparable
Terminal-Bench 2.1
Claude Opus 5—Hy4 preview85.4%SourceNot directly comparable
NL2Repo
Claude Opus 5—Hy4 preview58.9%SourceNot directly comparable
PostTrain Bench
Claude Opus 5—Hy4 preview35.6%SourceNot directly comparable
sweMarathon
Claude Opus 5—Hy4 preview31.9%SourceNot directly comparable
Reasoning
Knowledge
HLE
Claude Opus 564.7%SourceHy4 preview55.4%SourceClaude Opus 5 leads this result
HLE w/o tools
Claude Opus 556.3%SourceHy4 preview43.4%SourceClaude Opus 5 leads this result
HealthBench (raw)
Claude Opus 567.1%SourceHy4 preview—Not directly comparable
HealthBench (length-adjusted)
Claude Opus 557.8%SourceHy4 preview—Not directly comparable
HealthBench Professional
Claude Opus 559.8%SourceHy4 preview—Not directly comparable
HealthBench Professional (raw)
Claude Opus 573.4%SourceHy4 preview—Not directly comparable
BioMysteryBench (human-solvable)
Claude Opus 590.1%SourceHy4 preview—Not directly comparable
BioMysteryBench (human-difficult)
Claude Opus 549.4%SourceHy4 preview—Not directly comparable
SpatialBench Verified
Claude Opus 572.5%SourceHy4 preview—Not directly comparable
SingleCellBench
Claude Opus 560.6%SourceHy4 preview—Not directly comparable
ProteinGym Hard
Claude Opus 547.7%SourceHy4 preview—Not directly comparable
Protein Design
Claude Opus 542.5%SourceHy4 preview—Not directly comparable
Organic chemistry V2
Claude Opus 561.6%SourceHy4 preview—Not directly comparable
Protocols (troubleshooting)
Claude Opus 561.1%SourceHy4 preview—Not directly comparable
Protocols (understanding)
Claude Opus 578.4%SourceHy4 preview—Not directly comparable
GPQA
Claude Opus 5—Hy4 preview92.3%SourceNot directly comparable
GPQA-D
Claude Opus 5—Hy4 preview92.3%SourceNot directly comparable
Math
IMO 2026
Claude Opus 542/42SourceHy4 preview—Not directly comparable
RiemannBench (no tools)
Claude Opus 560.0%SourceHy4 preview—Not directly comparable
RiemannBench (tools)
Claude Opus 579.0%SourceHy4 preview—Not directly comparable
ArXivMath Jun. 2026 (no tools)
Claude Opus 590.8%SourceHy4 preview—Not directly comparable
ArXivMath Jun. 2026 (tools)
Claude Opus 591.3%SourceHy4 preview—Not directly comparable
Apex
Claude Opus 5—Hy4 preview74.2%SourceNot directly comparable
Multilingual
Multimodal
Chartography (no tools)
Claude Opus 529.6%SourceHy4 preview—Not directly comparable
Chartography (tools)
Claude Opus 583.0%SourceHy4 preview—Not directly comparable
BenchCAD Vision2Code (no tools)
Claude Opus 50.366SourceHy4 preview—Not directly comparable
BenchCAD Vision2Code (tools)
Claude Opus 50.821SourceHy4 preview—Not directly comparable
GDP.pdf (no tools)
Claude Opus 583.4%SourceHy4 preview—Not directly comparable
GDP.pdf (tools)
Claude Opus 585.5%SourceHy4 preview—Not directly comparable
OfficeQA
Claude Opus 578.1%SourceHy4 preview—Not directly comparable
OfficeQA Pro
Claude Opus 566.9%SourceHy4 preview66.2%SourceClaude Opus 5 leads this result
Frequently asked questions
Which is better, Claude Opus 5 or Hy4 preview?
Claude Opus 5 has the higher public score estimate, 83.06 versus 79.16, 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 5 or Hy4 preview?
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 5 or Hy4 preview?
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 5 or Hy4 preview?
Both models do not have comparable published API token rates, so this page does not name a universal price winner.
Which has the larger context window, Claude Opus 5 or Hy4 preview?
Both models list the same context window, 1M.
Related comparisons
Last updated August 28, 2026
Watch Claude Opus 5 vs Hy4 preview
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.