Coding
Directional only- Claude 3 Haiku
- 22.5
- Estimated · #181/183
- Claude Opus 5
- 75.6
- Supported · #3/183
- Basis
- BenchAlign lane · 0 vs 16 public rows
- Reading
- Directional only
Claude 3 Haiku vs Claude Opus 5
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.
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
Claude 3 Haiku
Claude 3 Haiku 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
Claude 3 Haiku
Claude 3 Haiku 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
Claude 3 Haiku is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
Claude 3 Haiku 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. Claude 3 Haiku does not fit this workload in one request. Claude 3 Haiku 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.
73 Claude Opus 5 only
- Shared results
- 0
- Claude 3 Haiku only
- 0
- Claude Opus 5 only
- 73
- Like-for-like categories
- 0 / 8
2 categories rest 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 3 Haiku
- 24.9
- Estimated · #178/181
- Claude Opus 5
- 82.1
- Supported · #3/181
- Basis
- BenchAlign lane · 0 vs 19 public rows
- Reading
- Directional only
Agentic
Not comparable- Claude 3 Haiku
- Not ranked
- Claude Opus 5
- 77.4
- Supported · #2/151
- Basis
- BenchAlign lane · 0 vs 19 public rows
- Reading
- Not comparable
Reasoning
Not comparable- Claude 3 Haiku
- 39.0
- Unranked · 2 rankable rows
- Claude Opus 5
- 75.6
- #13/22
- Basis
- Provisional lane · 0 vs 2 weighted rows
- Reading
- Not comparable
Math
Not comparable- Claude 3 Haiku
- Not ranked
- Claude Opus 5
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Multilingual
Not comparable- Claude 3 Haiku
- Not ranked
- Claude Opus 5
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
Multimodal
Not comparable- Claude 3 Haiku
- 25.3
- Unranked · 1 rankable row
- Claude Opus 5
- 88.7
- #2/48
- Basis
- Provisional lane · 0 vs 1 weighted rows
- Reading
- Not comparable
Instruction following
Not comparable- Claude 3 Haiku
- 41.3
- #101/120
- Claude Opus 5
- Not ranked
- Basis
- Provisional lane · 0 vs 0 weighted rows
- Reading
- Not comparable
| Category | Claude 3 Haiku | Claude Opus 5 | Basis | Reading |
|---|---|---|---|---|
| Coding | 22.5Estimated · #181/183 | 75.6Supported · #3/183 | Directional onlyBenchAlign lane · 0 vs 16 public rows | Directional only |
| Knowledge | 24.9Estimated · #178/181 | 82.1Supported · #3/181 | Directional onlyBenchAlign lane · 0 vs 19 public rows | Directional only |
| Agentic | Not ranked | 77.4Supported · #2/151 | Not comparableBenchAlign lane · 0 vs 19 public rows | Not comparable |
| Reasoning | 39.0Unranked · 2 rankable rows | 75.6#13/22 | Not comparableProvisional lane · 0 vs 2 weighted rows | Not comparable |
| Math | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Multilingual | Not ranked | Not ranked | Not comparableProvisional lane · 0 vs 0 weighted rows | Not comparable |
| Multimodal | 25.3Unranked · 1 rankable row | 88.7#2/48 | Not comparableProvisional lane · 0 vs 1 weighted rows | Not comparable |
| Instruction following | 41.3#101/120 | Not ranked | 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 3 Haiku
- $0.00088
- Fits in one request
- Claude Opus 5
- $0.0175
- Fits in one request
Claude 3 Haiku has the lower modeled cost
Costs use the listed standard API rates.
Repository review
50K fresh input + 3K output tokens
- Claude 3 Haiku
- $0.01625
- Fits in one request
- Claude Opus 5
- $0.325
- Fits in one request
Claude 3 Haiku 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 3 Haiku
- $0.0675
- Does not fit in one request
- Cached input priced at the published list-input rate
- Claude Opus 5
- $0.45
- Fits in one request
Claude 3 Haiku does not fit this workload in one request. Claude 3 Haiku 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 3 HaikuClaude Opus 5
Context window
Maximum documented context; output-token limits may be lower.
Claude 3 Haiku
200K
Claude Opus 5
API model ID
Claude 3 Haiku
Not sourced
Claude Opus 5
claude-opus-5
Anthropic model overviewCached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Claude 3 Haiku
Not published
Claude Opus 5
$0.5 per 1M cached input tokens
Claude API pricingDocumented inputs
Claude 3 Haiku
Not sourced
Claude Opus 5
text, image
Anthropic model overviewDocumented outputs
Claude 3 Haiku
Not sourced
Claude Opus 5
Provider availability
Claude 3 Haiku
Not sourced
Claude Opus 5
Generally Available · Claude API
Anthropic model overviewReasoning profile
Claude 3 Haiku
Non-Reasoning
Claude Opus 5
Reasoning
Weight access
Claude 3 Haiku
Proprietary
Claude Opus 5
Proprietary
License
Claude 3 Haiku
Proprietary
Claude Opus 5
Proprietary
Release date
Claude 3 Haiku
2024-03-01
Claude Opus 5
2026-07-24
If you already use one of these models
- Deployment change
- Both entries list Anthropic as the provider. Confirm endpoint, model ID, limits, and feature support before switching.
- 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: $0.01625 vs $0.325. Cache-heavy agent loop: $0.0675 vs $0.45.
- 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 evidence73 rows
Agentic
Terminal-Bench 3.0
Claude 3 Haiku—Claude Opus 542.7%SourceNot directly comparable
BrowseComp
Claude 3 Haiku—Claude Opus 590.8%SourceNot directly comparable
HLE w/ tools
Claude 3 Haiku—Claude Opus 564.7%SourceNot directly comparable
DeepSearchQA
Claude 3 Haiku—Claude Opus 595.0%SourceNot directly comparable
DRACO
Claude 3 Haiku—Claude Opus 588.6%SourceNot directly comparable
BrowseComp (10-agent, prerelease)
Claude 3 Haiku—Claude Opus 593.6%SourceNot directly comparable
OSWorld 2.0
Claude 3 Haiku—Claude Opus 570.6%SourceNot directly comparable
MCP Atlas
Claude 3 Haiku—Claude Opus 585.8%SourceNot directly comparable
MCP-Atlas claim coverage
Claude 3 Haiku—Claude Opus 589.1%SourceNot directly comparable
LAB all-pass (Anthropic harness)
Claude 3 Haiku—Claude Opus 523.58%SourceNot directly comparable
LAB criterion-pass (Anthropic harness)
Claude 3 Haiku—Claude Opus 593.74%SourceNot directly comparable
LAB all-pass (Harvey held-out)
Claude 3 Haiku—Claude Opus 511.7%SourceNot directly comparable
LAB criterion-pass (Harvey held-out)
Claude 3 Haiku—Claude Opus 594.1%SourceNot directly comparable
Toolathlon-Verified
Claude 3 Haiku—Claude Opus 580.6%SourceNot directly comparable
Toolathlon Verified Pass@3
Claude 3 Haiku—Claude Opus 587.0%SourceNot directly comparable
Toolathlon Verified Pass³
Claude 3 Haiku—Claude Opus 573.1%SourceNot directly comparable
Toolathlon Verified avg. turns
Claude 3 Haiku—Claude Opus 523.5 turnsSourceNot directly comparable
AutomationBench
Claude 3 Haiku—Claude Opus 526.0%SourceNot directly comparable
Terminal-Bench 2.1 (Vals)
Claude 3 Haiku—Claude Opus 584.6%SourceNot directly comparable
Coding
Bug Hunt Bench
Claude 3 Haiku—Claude Opus 527 fixesSourceNot directly comparable
SWE-bench Verified
Claude 3 Haiku—Claude Opus 596%SourceNot directly comparable
SWE-bench Pro
Claude 3 Haiku—Claude Opus 579.2%SourceNot directly comparable
SWE Multilingual
Claude 3 Haiku—Claude Opus 589.5%SourceNot directly comparable
SWE Multimodal
Claude 3 Haiku—Claude Opus 559.4%SourceNot directly comparable
deepSwe
Claude 3 Haiku—Claude Opus 568.8%SourceNot directly comparable
FrontierCode 1.1 Main
Claude 3 Haiku—Claude Opus 553.4%SourceNot directly comparable
FrontierCode 1.1 Extended
Claude 3 Haiku—Claude Opus 563.6%SourceNot directly comparable
FrontierSWE v2
Claude 3 Haiku—Claude Opus 552.0%SourceNot directly comparable
ProgramBench (episode 1)
Claude 3 Haiku—Claude Opus 583.0%SourceNot directly comparable
ProgramBench
Claude 3 Haiku—Claude Opus 593.0%SourceNot directly comparable
cursorBench32
Claude 3 Haiku—Claude Opus 570.0%SourceNot directly comparable
VulcanBench v3
Claude 3 Haiku—Claude Opus 587.0%SourceNot directly comparable
VulcanBench CII v1
Claude 3 Haiku—Claude Opus 596.4%SourceNot directly comparable
LiveCodeBench (Vals)
Claude 3 Haiku—Claude Opus 589.0%SourceNot directly comparable
SWE-bench (Vals)
Claude 3 Haiku—Claude Opus 597.0%SourceNot directly comparable
Reasoning
Knowledge
HLE
Claude 3 Haiku—Claude Opus 564.7%SourceNot directly comparable
HLE w/o tools
Claude 3 Haiku—Claude Opus 556.3%SourceNot directly comparable
HLE-Verified
Claude 3 Haiku—Claude Opus 554.4%SourceNot directly comparable
LABBench2
Claude 3 Haiku—Claude Opus 584.2%SourceNot directly comparable
HealthBench (raw)
Claude 3 Haiku—Claude Opus 567.1%SourceNot directly comparable
HealthBench (length-adjusted)
Claude 3 Haiku—Claude Opus 557.8%SourceNot directly comparable
HealthBench Professional
Claude 3 Haiku—Claude Opus 559.8%SourceNot directly comparable
HealthBench Professional (raw)
Claude 3 Haiku—Claude Opus 573.4%SourceNot directly comparable
BioMysteryBench (human-solvable)
Claude 3 Haiku—Claude Opus 590.1%SourceNot directly comparable
BioMysteryBench (human-difficult)
Claude 3 Haiku—Claude Opus 549.4%SourceNot directly comparable
SpatialBench Verified
Claude 3 Haiku—Claude Opus 572.5%SourceNot directly comparable
SingleCellBench
Claude 3 Haiku—Claude Opus 560.6%SourceNot directly comparable
ProteinGym Hard
Claude 3 Haiku—Claude Opus 547.7%SourceNot directly comparable
Protein Design
Claude 3 Haiku—Claude Opus 542.5%SourceNot directly comparable
Organic chemistry V2
Claude 3 Haiku—Claude Opus 561.6%SourceNot directly comparable
Protocols (troubleshooting)
Claude 3 Haiku—Claude Opus 561.1%SourceNot directly comparable
Protocols (understanding)
Claude 3 Haiku—Claude Opus 578.4%SourceNot directly comparable
GPQA Diamond (Vals)
Claude 3 Haiku—Claude Opus 593.4%SourceNot directly comparable
MMLU-Pro (Vals)
Claude 3 Haiku—Claude Opus 591.6%SourceNot directly comparable
Math
IMO 2026
Claude 3 Haiku—Claude Opus 542/42SourceNot directly comparable
RiemannBench (no tools)
Claude 3 Haiku—Claude Opus 560.0%SourceNot directly comparable
RiemannBench (tools)
Claude 3 Haiku—Claude Opus 579.0%SourceNot directly comparable
ArXivMath Jun. 2026 (no tools)
Claude 3 Haiku—Claude Opus 590.8%SourceNot directly comparable
ArXivMath Jun. 2026 (tools)
Claude 3 Haiku—Claude Opus 591.3%SourceNot directly comparable
Multilingual
Multimodal
Chartography (no tools)
Claude 3 Haiku—Claude Opus 529.6%SourceNot directly comparable
Chartography (tools)
Claude 3 Haiku—Claude Opus 583.0%SourceNot directly comparable
BenchCAD Vision2Code (no tools)
Claude 3 Haiku—Claude Opus 50.366SourceNot directly comparable
BenchCAD Vision2Code (tools)
Claude 3 Haiku—Claude Opus 50.821SourceNot directly comparable
GDP.pdf (no tools)
Claude 3 Haiku—Claude Opus 583.4%SourceNot directly comparable
GDP.pdf (tools)
Claude 3 Haiku—Claude Opus 585.5%SourceNot directly comparable
OfficeQA
Claude 3 Haiku—Claude Opus 578.1%SourceNot directly comparable
OfficeQA Pro
Claude 3 Haiku—Claude Opus 566.9%SourceNot directly comparable
Frequently asked questions
Which is better, Claude 3 Haiku or Claude Opus 5?
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 3 Haiku or Claude Opus 5?
Claude Opus 5 scores higher for coding on the public lane, 75.6 to 22.5. Claude 3 Haiku is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like. BenchLM does not name a winner for coding from a directional reading; read the shared benchmark rows directly and test the models on the same task set.
Which is better for agentic tasks, Claude 3 Haiku or Claude Opus 5?
Claude 3 Haiku is not ranked on the public lane for agentic tasks, so no winner is named for agentic tasks.
Which costs less, Claude 3 Haiku or Claude Opus 5?
For the stated presets, chat costs $0.00088 on Claude 3 Haiku and $0.0175 on Claude Opus 5; repository review costs $0.01625 and $0.325; the cache-heavy agent loop costs $0.0675 and $0.45. Claude 3 Haiku does not fit this workload in one request. Claude 3 Haiku has no published cached-input rate, so cached tokens use its listed input rate.
Which has the larger context window, Claude 3 Haiku or Claude Opus 5?
Claude Opus 5 has the larger documented context window: 1M, compared with 200K.
Related comparisons
Last updated September 4, 2026
Watch Claude 3 Haiku vs Claude Opus 5
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.