Claude Haiku Latest
Not recorded first token, 1.6 s full answer, 138.5 tokens per second, 0.20 credits, 100% success rate. Median of 6 checks, Oct 18, 2026 to Nov 17, 2026.
Anthropic: Claude Haiku 4.5 uses 2.4x fewer credits per answer. Median results from Krater's daily 120-word prompt over 30 days, including speed and credits per answer.
Claude Haiku 4.5 vs Claude Sonnet 4.6: strengths, best use cases and how to run both side by side. The page also lists context windows, token pricing and more comparisons.
Run Claude Haiku 4.5 and Claude Sonnet 4.6 on the same prompt in the Krater Model Arena, then keep the winner in your workflow. One subscription covers both, from $49 a month.
Open Krater chat · Claude Haiku 4.5 on Krater · Claude Sonnet 4.6 on Krater
The newest and fastest model in the Claude family. Claude Haiku 4.5 delivers impressive capabilities at lightning speed, perfect for real-time applications and high-volume usage.
The latest iteration of Claude Sonnet, pushing the boundary of what mid-tier models can achieve. Claude Sonnet 4.6 matches previous Opus-level performance at Sonnet pricing.
Speed and credit use for the fixed 120-word prompt.
| Measure | Anthropic: Claude Haiku 4.5 | Anthropic: Claude Sonnet 4.6 |
|---|---|---|
| First token | 967 ms | 647 ms |
| Full answer time | 2.1 s | 3.9 s |
| Tokens per second | 75.9 | 40.3 |
| Credits per answer | 0.20 credits | 0.48 credits |
| Success rate | 97% | 97% |
Median of 29 successful checks out of 30 for Anthropic: Claude Haiku 4.5 and 29 successful checks out of 30 for Anthropic: Claude Sonnet 4.6, Oct 18, 2026 to Nov 17, 2026. Same prompt every day. How we test: methodology. Model pricing.
Not recorded first token, 1.6 s full answer, 138.5 tokens per second, 0.20 credits, 100% success rate. Median of 6 checks, Oct 18, 2026 to Nov 17, 2026.
2.2 s first token, 3.2 s full answer, 68.3 tokens per second, 0.45 credits, 100% success rate. Median of 6 checks, Oct 18, 2026 to Nov 17, 2026.
Anthropic: Claude Sonnet 4.6 starts answering 1.5x faster.
Anthropic: Claude Haiku 4.5 uses 2.4x fewer credits per answer.
Claude Haiku Latest is the newer model in this family on Krater. Claude Sonnet Latest is the newer model in this family on Krater.
Our tested head-to-heads run the same prompt set through both models in the Model Arena, then rank answers blind. Read the testing methodology, or browse every comparison.