Claude Haiku 4.5 vs Claude Sonnet 4.5

Anthropic against Anthropic, the same lab's consecutive models. Prices, scores and speed below are live.

  • Claude Haiku 4.5 costs 67% less on blended price ($3.00 against $9.00 per 1M tokens, 3 input to 1 output).
  • Claude Sonnet 4.5 scores higher on the Artificial Analysis Intelligence Index: 21.2 against 7.4.
  • On the 40 benchmarks both have scores on, Claude Haiku 4.5 leads on 1 and Claude Sonnet 4.5 on 39.
  • Claude Sonnet 4.5 reads more context: 1M tokens against 200K.
  • Claude Haiku 4.5 is newer: listed Oct 15, 2025, against Sep 29, 2025.

Side by side

Claude Haiku 4.5Claude Sonnet 4.5
LabAnthropicAnthropic
ListedOct 15, 2025Sep 29, 2025
Input per 1M tokens$1.50$4.50
Output per 1M tokens$7.50$22.50
Cached input per 1M$0.15$0.45
Blended (3:1) per 1M$3.00$9.00
Context window200K1M
Max output64K64K
Knowledge cutoff-2025-01-31
Intelligence index7.421.2
Coding index43.952.1
ReasoningYesYes
Reads imagesYesYes
Tool useYesYes
Structured outputYesYes

Every benchmark both are scored on

Each model's best score per benchmark, with its source. Scores from different sources or settings are not always comparable; each links to where it was published.

BenchmarkClaude Haiku 4.5Claude Sonnet 4.5
AIME66.7%Epoch AI77.8%Epoch AI
ARC-AGI-147.7%Epoch AI63.7%Epoch AI
ARC-AGI-24%Epoch AI13.6%Epoch AI
Artificial Analysis Agentic Index8Artificial Analysis and OpenRouter15.8Artificial Analysis and OpenRouter
Artificial Analysis Coding Index43.9Artificial Analysis and OpenRouter52.1Artificial Analysis and OpenRouter
Artificial Analysis Intelligence Index7.4Artificial Analysis and OpenRouter21.2Artificial Analysis and OpenRouter
Chess Puzzles8%Epoch AI10%Epoch AI
DeepResearchBench (FutureSearch)45.5Epoch AI52.6Epoch AI
Design Arena: 3D1057 EloDesign Arena1143 EloDesign Arena
Design Arena: ASCII Art1137 EloDesign Arena1202 EloDesign Arena
Design Arena: Code1109 EloDesign Arena1177 EloDesign Arena
Design Arena: Data Viz1122 EloDesign Arena1165 EloDesign Arena
Design Arena: Game Dev1091 EloDesign Arena1160 EloDesign Arena
Design Arena: SVG1017 EloDesign Arena1107 EloDesign Arena
Design Arena: UI Component1093 EloDesign Arena1165 EloDesign Arena
Design Arena: Website1118 EloDesign Arena1184 EloDesign Arena
DTBench73.6%Epoch AI83.2%Epoch AI
FrontierMath5.9%Epoch AI9.3%Epoch AI
FrontierMath Tier 4 (v1, archived)2.1%Epoch AI4.2%Epoch AI
GPQA Diamond73%Artificial Analysis and OpenRouter81.7%Artificial Analysis and OpenRouter
GPQA Main60.5%Stanford HELM68.6%Stanford HELM
IFEval80.1%Stanford HELM85%Stanford HELM
LMArena (Coding)1453 EloLMArena1485 EloLMArena
LMArena (Overall)1397 EloLMArena1439 EloLMArena
LMArena: Creative Writing1371 EloLMArena1442 EloLMArena
LMArena: Hard Prompts1420 EloLMArena1462 EloLMArena
LMArena: Longer Query1426 EloLMArena1476 EloLMArena
LMArena: Math1430 EloLMArena1445 EloLMArena
LMArena: Multi-Turn1409 EloLMArena1465 EloLMArena
LMCA30.9Epoch AI38.8Epoch AI
MATH (Level 5)96.4%Epoch AI97.7%Epoch AI
MMLU-Pro77.7%Stanford HELM86.9%Stanford HELM
Omni-MATH56.1%Stanford HELM55.3%Stanford HELM
SciCode42.2%Epoch AI45.7%Epoch AI
SimpleQA Verified12.6%Epoch AI30.7%Epoch AI
tau-bench (Airline)66.2%Artificial Analysis and OpenRouter72.5%Artificial Analysis and OpenRouter
Terminal-Bench35.5%Epoch AI46.5%Epoch AI
WebDev Arena1286 EloEpoch AI1391 EloEpoch AI
WeirdML45.4%Epoch AI47.7%Epoch AI
WildBench8.6Stanford HELM8.7Stanford HELM

More comparisons

Where these figures come from

Prices are GenMagic's own, what the studio and the API charge, recomputed from each provider's price list every few minutes. The indices are the Artificial Analysis indices, and every benchmark score links to its source, re-read daily. Measured speed is GenMagic's own generation records, in aggregate. See Anthropic models, Anthropic models, every LLM's price and our sources.