Claude Sonnet 4.5
Not tested yet
Head-to-head comparison
Claude Sonnet 4.5 vs. Claude Haiku 4.5
Compare published benchmark results from matching versions and cohorts, API pricing, and technical specifications. Any documented test setup differences remain visible.
- shared result series
- 4
- distinct benchmarks
- 2
- latest retrieval
- 2026-08-06
Change model selection
Specifications
General specifications
Release date, availability, context window, architecture, and published parameter counts at a glance. Undisclosed details remain marked as unavailable.
| Attribute | Claude Sonnet 4.5 | Claude Haiku 4.5 |
|---|---|---|
| Provider | Anthropic | Anthropic |
| Released | Sep 2025 | Oct 2025 |
| Availability | Active | Active |
| Model type | Proprietary | Proprietary |
| Parameters | Not published | Not published |
| Architecture | Not published | Not published |
| Context window | 200,000 tokens | 200,000 tokens |
| Knowledge cutoff | January 2025 | February 2025 |
| Technical sources | Model, Context | Model |
Pricing
API pricing compared
The table compares official standard API rates per 1M tokens, including published cache pricing. Batch, fast, priority, regional surcharges, tool calls, and cloud platform pricing are excluded.
| Price type | Claude Sonnet 4.5 | Claude Haiku 4.5 |
|---|---|---|
| API input | $3 | $1 |
| API output | $15 | $5 |
| Cache read | $0.3 | $0.1 |
| Cache write (5 min.) | $3.75 | $1.25 |
| Cache write (1 hr.) | $6 | $2 |
| Price verified | 07/30/2026 Source | 07/30/2026 Source |
Claude Sonnet 4.5Claude Haiku 4.5
API inputUSD per 1M tokens, base tier
Claude Sonnet 4.5$3
Claude Haiku 4.5Lower price$1
API outputUSD per 1M tokens, base tier
Claude Sonnet 4.5$15
Claude Haiku 4.5Lower price$5
Cache readUSD per 1M tokens, base tier
Claude Sonnet 4.5$0.3
Claude Haiku 4.5Lower price$0.1
Key to the marks:better valuebetter, but not conclusivetie
Pricing by provider
Direct developer pricing and provider endpoint tariffs routed through OpenRouter are listed separately. Prices are in USD per 1M tokens. Context-dependent direct tiers plus regional, Flex, Priority, and other OpenRouter tariffs remain individually identifiable.
| Model | Provider | Purchase route and tariff | Input per 1M tokens | Output per 1M tokens | Source |
|---|---|---|---|---|---|
| Claude Sonnet 4.5 | Anthropic | Direct from developerStandard API | Lowest listed price: $3 | Lowest listed price: $15 | Source07/30/2026 |
| Amazon Bedrock | Via OpenRouteramazon-bedrock | Lowest listed price: $3 | Lowest listed price: $15 | Source09/12/2026 | |
| Amazon Bedrock | Via OpenRouteramazon-bedrock/eu-west-1 | $3.3 | $16.5 | Source09/12/2026 | |
| Anthropic | Via OpenRouteranthropic | Lowest listed price: $3 | Lowest listed price: $15 | Source09/12/2026 | |
| Azure | Via OpenRouterazure/global | Lowest listed price: $3 | Lowest listed price: $15 | Source09/12/2026 | |
| Claude Platform on AWS | Via OpenRouterclaude-on-aws | Lowest listed price: $3 | Lowest listed price: $15 | Source09/12/2026 | |
| Via OpenRoutergoogle-vertex/global | Lowest listed price: $3 | Lowest listed price: $15 | Source09/12/2026 | ||
| Via OpenRoutergoogle-vertex/us-east5 | $3.3 | $16.5 | Source09/12/2026 | ||
| Claude Haiku 4.5 | Anthropic | Direct from developerStandard API | Lowest listed price: $1 | Lowest listed price: $5 | Source07/30/2026 |
| Amazon Bedrock | Via OpenRouteramazon-bedrock/eu-west-1 | $1.1 | $5.5 | Source09/12/2026 | |
| Amazon Bedrock | Via OpenRouteramazon-bedrock/global | Lowest listed price: $1 | Lowest listed price: $5 | Source09/12/2026 | |
| Amazon Bedrock | Via OpenRouteramazon-bedrock/us | $1.1 | $5.5 | Source09/12/2026 | |
| Anthropic | Via OpenRouteranthropic | Lowest listed price: $1 | Lowest listed price: $5 | Source09/12/2026 | |
| Azure | Via OpenRouterazure/global | Lowest listed price: $1 | Lowest listed price: $5 | Source09/12/2026 | |
| Via OpenRoutergoogle-vertex/europe | $1.1 | $5.5 | Source09/12/2026 | ||
| Via OpenRoutergoogle-vertex/global | Lowest listed price: $1 | Lowest listed price: $5 | Source09/12/2026 | ||
| Via OpenRoutergoogle-vertex/us-east5 | $1.1 | $5.5 | Source09/12/2026 |
Performance
Shared benchmarks
Only published results with the same benchmark version, task, metric, and comparison cohort are paired. Different reasoning levels, agents, harnesses, or output limits appear directly in the table.
Capability profile from matched benchmarks
Each axis averages directly matched benchmark families. A value of 100 means the stronger value within this pair, not a universal quality score.
Not enough directly matched categories for a reliable radar
Only 1 of at least 5 categories have comparable, direct model measurements. The benchmark tables below show the available evidence without filling missing axes.
Claude Sonnet 4.5Claude Haiku 4.5
ScienceAgentBenchAccuracy. Higher is better. No documented setup difference
Claude Sonnet 4.5Winner29.41 %
Claude Haiku 4.518.63 %
SWE-bench Verified MiniAccuracy. Higher is better. No documented setup difference
Claude Sonnet 4.5Winner34 %
Claude Haiku 4.524 %
ScienceAgentBenchAccuracy. Higher is better. No documented setup difference
Claude Sonnet 4.5Winner30.39 %
Claude Haiku 4.523.53 %
SWE-bench Verified MiniAccuracy. Higher is better. No documented setup difference
Claude Sonnet 4.540 %
Claude Haiku 4.5Winner44 %
Key to the marks:better valuebetter, but not conclusivetie
| Benchmark and task | Claude Sonnet 4.5 | Claude Haiku 4.5 | Source |
|---|---|---|---|
ScienceAgentBenchAccuracyTest detailsNo documented difference in the test setupVersion: verified 102-task setMetric: accuracyScoring: Higher is betterSettings: Agent: SAB Self-Debug, Harness: SAB Self-DebugSAB Self-Debug on 102 scientific tasks. Total cost is divided by 102 and excludes caching benefits according to HAL. | Winner29.41 % | 18.63 % | Princeton HALSource detailsData as of: 08/06/2026 |
SWE-bench Verified MiniAccuracyTest detailsNo documented difference in the test setupVersion: 50-task random verified subsetMetric: accuracyScoring: Higher is betterSettings: Agent: HAL Generalist, Harness: HAL GeneralistHAL Generalist on the same random 50-task subset. Total cost is divided by 50. | Winner34 % HAL reports min-max ranges for repeated runs. The central value is the reported mean, not a 95% confidence interval. | 24 % HAL reports min-max ranges for repeated runs. The central value is the reported mean, not a 95% confidence interval. | Princeton HALSource detailsData as of: 08/05/2026 |
ScienceAgentBenchAccuracyTest detailsNo documented difference in the test setupVersion: verified 102-task setMetric: accuracyScoring: Higher is betterSettings: Reasoning: high, Agent: SAB Self-Debug, Harness: SAB Self-DebugSAB Self-Debug on 102 scientific tasks. Total cost is divided by 102 and excludes caching benefits according to HAL. | Winner30.39 % | 23.53 % | Princeton HALSource detailsData as of: 08/06/2026 |
SWE-bench Verified MiniAccuracyTest detailsNo documented difference in the test setupVersion: 50-task random verified subsetMetric: accuracyScoring: Higher is betterSettings: Reasoning: high, Agent: HAL Generalist, Harness: HAL GeneralistHAL Generalist on the same random 50-task subset. Total cost is divided by 50. | 40 % HAL reports min-max ranges for repeated runs. The central value is the reported mean, not a 95% confidence interval. | Winner44 % HAL reports min-max ranges for repeated runs. The central value is the reported mean, not a 95% confidence interval. | Princeton HALSource detailsData as of: 08/05/2026 |
Key to the marks:better valuebetter, but not conclusivetie | |||
Our model tests
Community tests you can inspect
Compare how the models respond to the same prompt. Each test shows the first attempt, with no subsequent fixes to the generated code. These results do not contribute to an overall score.
Pelican on a bicycle
A community classic for free-form SVG drawing.
Claude Haiku 4.5
Not tested yet
Prompt and test conditions
Generate an SVG of a pelican riding a bicycle
Token limit including reasoning: 8,192 tokens
Task origin (Simon Willison)How to read this comparison
Winning one benchmark is not an overall verdict
Your workload, budget, and required context length matter most. A coding benchmark says little about visual understanding or agent performance.
The standard price table uses direct developer rates. The provider comparison labels OpenRouter endpoints, regions, and special tariffs separately.
Every result links to its measurement source and retrieval date. Labels distinguish vendor reports, official benchmark leaderboards, and independent evaluations. Disputed or archived results are excluded.
More matchups
Related LLM comparisons
Compare Claude Sonnet 4.5 and Claude Haiku 4.5 with other leading models using the same data and benchmark logic.