Claude 4.5 Haiku (Reasoning) vs Claude 4.5 Sonnet (Reasoning)

Comparing 2 AI models · 5 benchmarks · Anthropic

Most Affordable
Anthropic logo
Claude 4.5 Haiku (Reasoning)
$1.00/1M
Highest Intelligence
Anthropic logo
Claude 4.5 Sonnet (Reasoning)
83.4% GPQA
Best for Coding
Anthropic logo
Claude 4.5 Sonnet (Reasoning)
49.8 Coding Index
Price Difference
3.0x
input cost range

Composite Indices

Intelligence, Coding, Math

Standard Benchmarks

Academic and industry benchmarks

Benchmark Winners

5 tests
Anthropic logo

Claude 4.5 Haiku (Reasoning)

0

No clear wins

Anthropic logo

Claude 4.5 Sonnet (Reasoning)

5
  • GPQA
  • MMLU Pro
  • HLE
  • LiveCodeBench
  • AIME 2025
Metric
Anthropic logo Claude 4.5 Haiku (Reasoning)
Anthropic
Anthropic logo Claude 4.5 Sonnet (Reasoning)
Anthropic
Pricing
Per 1M tokens
Input Cost $1.00/1M $3.00/1M
Output Cost $5.00/1M $15.00/1M
Blended Cost 3:1 input/output ratio
$2.00/1M $6.00/1M
Specifications
Organization Model creator
Anthropic Anthropic
Release Date Launch date
Oct 15, 2025 Sep 29, 2025
Performance & Speed
Throughput Output speed
51.8 tok/s 72.5 tok/s
Time to First Token (TTFT) Initial response delay
1087ms 2014ms
Latency Time to first answer token
39706ms 29615ms
Composite Indices
Intelligence Index Overall reasoning capability
54.6 62.7
Coding Index Programming ability
43.4 49.8
Math Index Mathematical reasoning
83.7 88.0
Standard Benchmarks
GPQA Graduate-level reasoning
67.2% 83.4%
MMLU Pro Advanced knowledge
76.0% 87.5%
HLE Hard language evaluation
9.7% 17.3%
LiveCodeBench Real-world coding tasks
61.5% 71.4%
MATH 500 Mathematical problems
AIME 2025 Advanced math competition
83.7% 88.0%
AIME (Original) Math olympiad problems
SciCode Scientific code generation
43.3% 44.7%
LCR Code review capability
70.3% 65.7%
IFBench Instruction-following
54.3% 57.3%
TAU-bench v2 Tool use & agentic tasks
54.7% 78.1%
TerminalBench Hard CLI command generation
25.5% 33.3%

Key Takeaways

Claude 4.5 Haiku (Reasoning) offers the best value at $1.00/1M, making it ideal for high-volume applications and cost-conscious projects.

Claude 4.5 Sonnet (Reasoning) leads in reasoning capabilities with a 83.4% GPQA score, excelling at complex analytical tasks and problem-solving.

Claude 4.5 Sonnet (Reasoning) achieves a 49.8