benchrPricing breakdown
| Tier | Rate / 1M tokens |
|---|---|
| Standard input | $1.00 |
| Standard output | $5.00 |
| Cached input | $0.10 |
| Context window | 200,000 tokens |
| Max output | 8,192 tokens |
What 145 tok/s actually means
At 145 tokens per second, Haiku 4.5 generates a 500-word response in roughly 3.5 seconds. Claude Sonnet 4.6 at approximately 85 tok/s takes about 6 seconds for the same response. For customer-facing applications, this 40% latency improvement is perceptible and directly affects user experience. For batch pipelines, it means processing the same workload in 60% of the time — or running 67% more volume within the same time window.
The speed advantage compounds in streaming applications where first-token latency matters: Haiku typically delivers the first token faster than larger models, which improves perceived responsiveness even before throughput becomes relevant.
Where the 73.3% SWE-bench score creates limits
Haiku 4.5's 73.3% SWE-bench is 6 points below Sonnet 4.6's 79.6%. In practical terms, this means Haiku resolves roughly 6 fewer issues per 100 than Sonnet on real-world coding tasks. For simple, well-specified tasks — fixing a single function, explaining code, generating boilerplate — the gap is invisible. For multi-file refactors, architecture-level changes, or debugging complex logic, Haiku's lower ceiling becomes measurable. The 8,192 max output limit also constrains it for long code generation tasks.
Cost scenarios
At 50M input + 10M output per month — a moderate production volume: Haiku 4.5 costs $50 + $50 = $100/month. Claude Sonnet 4.6 at the same volume: $150 + $150 = $300/month — 3× the cost. With 90% cache hit on Haiku: approximately $9.50 + $50 = $59.50/month. For a product routing 10,000 user queries per day through a classification step, Haiku at $100/month versus Sonnet at $300/month is a $200/month cost difference that scales directly with volume.
Use-case fit
Best for: High-volume classification and intent detection; customer service first-response routing; real-time chat with latency constraints; extraction from structured documents; any pipeline where cost-per-call and throughput are primary constraints and 73.3% SWE-bench quality is sufficient.
Skip if: Your tasks require complex multi-step code generation or architecture reasoning — the 6-point SWE-bench gap to Sonnet is measurable here. Also skip for long output tasks; the 8,192 output limit will force response truncation.
Decision checklist
Before choosing Haiku over Sonnet: run a sample of your actual tasks on both models and check for quality degradation in your specific failure modes. If your task set is pure classification or extraction with structured outputs, Haiku is likely sufficient. If it involves code generation or open-ended reasoning, measure pass rates on both before committing.
Check your required output length: if you routinely need responses over 5,000 tokens, Haiku's 8K output limit is a hard constraint — you'll need to split responses or use Sonnet instead.