Claude Haiku 4.5 vs Claude Sonnet 4
Wins 3 of 7 areas
Coding · Agents · Long documents
Wins 3 of 7 areas
Reasoning · Images and charts · Following instructions
The two are evenly matched.Claude Haiku 4.5 is better at coding and agents; Claude Sonnet 4 at reasoning and following instructions.
Scores updated · 41 tests both models report · How we compare
Where each one wins
Tests won in each of the seven areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- ReasoningHard problems that need careful thinking02Claude Sonnet 42 of 4 tests · 2 ties
- Following instructionsDoing exactly what it is asked01Claude Sonnet 41 of 2 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video01Claude Sonnet 41 of 1 test
- CodingWriting and fixing software32Claude Haiku 4.53 of 6 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own21Claude Haiku 4.52 of 3 tests
- Long documentsFinding answers in very long texts10Claude Haiku 4.51 of 1 test
- FactsGetting facts right instead of making them up11Even1 each · 1 tie
Images and charts and long documents rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
Claude Haiku 4.5 costs 67% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Claude Sonnet 4 pulls ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+10.5points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+7.4points ahead
- Keeps track of context across a multi-turn chatMulti-Challenge+6.6points ahead
Where Claude Haiku 4.5 pulls ahead
- Fixes real GitHub issues in many programming languagesSWE-bench Multilingual+10.5points ahead
- Completes tasks by operating a computer desktopOSWorld-Verified+8.5points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+7.9points ahead
Every test, side by side
All 41 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Haiku 4.5
- SWE-bench MultilingualClaude Haiku 4.5 by 10.567.456.9+10.5
- Terminal-Bench 2.1Claude Haiku 4.5 by 7.944.236.3+7.9
- Terminal-Bench HardClaude Sonnet 4 by 3.827.331.1+3.8
- SWE-bench ProClaude Sonnet 4 by 3.239.542.7+3.2
- SciCodeClaude Haiku 4.5 by 2.242.240+2.2
- SWE-bench Verifiedtie73.372.7tie
AgentsClaude Haiku 4.5
- OSWorld-VerifiedClaude Haiku 4.5 by 8.550.742.2+8.5
- τ-Bench V3 · BankingClaude Sonnet 4 by 7.49.316.7+7.4
- GDPValClaude Haiku 4.5 by 2.111.89.7+2.1
ReasoningClaude Sonnet 4
- GPQA DiamondClaude Sonnet 4 by 10.567.277.7+10.5
- ARC-AGI-2Claude Sonnet 4 by 1.945.9+1.9
- CritPttie00.3tie
- Humanity's Last Examtie10.410.7tie
FactsEven
- AA-Omniscience · AccuracyClaude Sonnet 4 by 4.71822.7+4.7
- AA-Omniscience · Non-hallucinationClaude Haiku 4.5 by 1.872.770.9+1.8
- Vectara HHEM hallucination ratelower is bettertie9.810.3tie
Images and chartsClaude Sonnet 4
- MMMU-ProClaude Sonnet 4 by 3.258.661.8+3.2
Following instructionsClaude Sonnet 4
- Multi-ChallengeClaude Sonnet 4 by 6.650.557.1+6.6
- IFBenchtie54.354.7tie
Other results21 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- AIME 2025Claude Haiku 4.5 by 10.280.770.5+10.2
- τ²-Bench Telecom (AA run)Claude Sonnet 4 by 9.954.764.6+9.9
- τ²-Bench (Retail)Claude Haiku 4.5 by 8.283.275+8.2
- Tau2 airlineClaude Haiku 4.5 by 8.163.655.5+8.1
- ARC-AGI-1Claude Haiku 4.5 by 7.747.740+7.7
- AA Agentic IndexClaude Sonnet 4 by 7.310.317.6+7.3
- Artificial Analysis Coding IndexClaude Haiku 4.5 by 6.343.937.6+6.3
- Terminal-BenchClaude Haiku 4.5 by 5.54135.5+5.5
- AIR-Bench 2024Claude Haiku 4.5 by 4.993.288.3+4.9
- AA-OmniscienceClaude Sonnet 4 by 4.6-4.40.2+4.6
- bbqClaude Sonnet 4 by 3.892.896.6+3.8
- MMMLUClaude Sonnet 4 by 3.58386.5+3.5
- XSTestClaude Sonnet 4 by 3.493.296.6+3.4
- vectara_avg_summary_lengthClaude Sonnet 4 by 30.7 rating points115.1145.8+30.7 rating
- anthropic_red_teamClaude Sonnet 4 by 2.396.999.2+2.3
- frontiermath_tier_4_v1Claude Haiku 4.5 by 2.12.10+2.1
- HarmBenchClaude Sonnet 4 by 2.195.998+2.1
- AA IntelligenceClaude Sonnet 4 by 216.918.9+2
- simple_safety_testsClaude Sonnet 4 by 1.298.8100+1.2
- vectara_answer_ratetie99.598.6tie
- vectara_factual_consistencytie90.289.7tie
Questions people ask
Which is better, Claude Haiku 4.5 or Claude Sonnet 4?
Claude Haiku 4.5 and Claude Sonnet 4 each win three of the seven areas where both have results. Claude Haiku 4.5 wins coding, agents and long documents; Claude Sonnet 4 wins reasoning, images and charts and following instructions. Claude Haiku 4.5 costs 67% less. They are level on facts.
Which is better for coding?
Claude Haiku 4.5. It wins 3 of the 6 coding tests both models report; Claude Sonnet 4 wins 2, and 1 is a tie.
Which is cheaper?
Claude Haiku 4.5 costs $1.00 per million input tokens and $5.00 per million output tokens; Claude Sonnet 4 costs $3.00 and $15.00. That makes Claude Haiku 4.5 about 67% cheaper for the same work.
How do you compare the two?
We use the 41 benchmark tests both models have published scores on. The verdict counts the 20 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 21 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.