‹ PublicAI Index
The LLM benchmark aggregator.
Claude Haiku 5.5
Anthropic
Strongest in Document design (#6 of 47), weakest in Core abilities (#168 of 225). Above par in 19 of 25 scopes. Among the models it meets almost everywhere, it finishes behind Claude Sonnet 5.5 and Claude Fable 5.1 and ahead of Kimi K3 and Gemini 3.6 Flash.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Agents61.1−4.8#11/2893/6
Long-horizon work60.3−13.7#9/471/1
Knowledge work66.6−5.2#12/1982/2
Web research57.8−5.9#13/481/2
Computer use58−9.9#20/801/2
Coding56.9−9.5#24/1763/6
Agentic coding57.2−9.8#22/1693/5
Professional54.1−9#38/1872/2
Document design62.1−5#6/471/1
Cybersecurity59.6−4.4#6/461/1
Medical60.6−5.9#7/1491/1
Biology research50.9−20.2#12/271/1
IT operations43.5−30.3#16/311/1
Finance54.2−8.3#61/1611/1
Legal48.4−18#94/1621/1
Reasoning54.9−9.9#46/1894/5
Mathematics58.4−5.6#17/1492/2
Reasoning51.2−15.5#61/1432/3
Knowledge50.8−12#100/1901/3
Document parsing50.9−12.8#44/761/1
Core abilities44.6−20.6#168/2253/4
General intelligence59.3−9.5#32/2142/3
Instruction following47.4−26.4#37/621/1
Data analysis26−39.9#46/471/1
Language30.8−35.8#58/621/1
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Claude Haiku 5.5, left for the other.
§ 3 · Sources
Where the numbers come from
8 publications, 37 figures. Every one links to the page it was read from.
Artificial Analysis Intelligence Index 43
GDPval-AA 1618
AA-Briefcase 1577
ARC-AGI-2 61.3%
LiveBench 75.6LiveBench · Agentic 57.6LiveBench · Reasoning 82.4LiveBench · Coding 91.4LiveBench · Browser Use 80LiveBench · Data Analysis 53.1LiveBench · Research 81.9LiveBench · Design 83.1
LiveBench 1.0 72.1LiveBench 1.0 · Reasoning 81.2LiveBench 1.0 · Coding 76.4LiveBench 1.0 · Agentic Coding 51.4LiveBench 1.0 · Mathematics 92.8LiveBench 1.0 · Data Analysis 72.8LiveBench 1.0 · Language 63.5LiveBench 1.0 · Instruction Following 66.5
Vals · Legal Research Bench 43.27%Vals · Harvey Legal Agent Benchmark 1.25%Vals · Finance Agent 54.14%Vals · MedCode 51.54%Vals · MedScribe 89.54%Vals · BioMysteryBench 70.37%Vals · CyberBench 75.48%Vals · SRE Bench 5.72%Vals · Vibe Code Bench 90.44% (OpenHands)Vals · Code Migration 44.57%Vals · ProofBench 89%
ParseBench · Overall 57.4%ParseBench · Tables 86.85%ParseBench · Charts 10.13%ParseBench · Content faithfulness 90.24%ParseBench · Semantic formatting 65.64%ParseBench · Visual grounding 34.14%
Badge
[](https://publicai.io/model-index/m/claude-haiku-5-5)