Banja Lab / Benchmarks

claude-opus-4-8

Low reasoningAgent harness

claude-opus-4-8 (Anthropic) scored 81.8% composite across 87 tasks - code, UI, full websites, SVG, marketing pages, dashboards, animations, and Australian legal and accounting. Graded by execution, and the visual builds by a cross-family vision panel (leave-one-family-out). Run on 2026-06-24.

Composite

81.8%

95% CI 74.4% - 88.7%

Est. cost

$3.11

USD

Tokens

163,926

generation

Wall clock

8 min

end to end

Gate pass

-

validated

Separation

-

good vs bad

By domain

Composite score per domain, weakest first. Judge is the vision model’s read, shown for the visual domains.

Websites

15 tasks

53.8%

composite

svg-scene

1 task

62.5%

judge

62.5%

composite

Programming

6 tasks

66.7%

composite

Animation

2 tasks

69.8%

judge

69.8%

composite

Marketing page

1 task

70.8%

judge

70.8%

composite

Auth screen

1 task

70.8%

judge

70.8%

composite

Australian law

5 tasks

80.0%

composite

Dashboard

1 task

83.3%

judge

83.3%

composite

Australian accounting

23 tasks

89.9%

composite

SVG and graphics

13 tasks

94.2%

composite

UI components

19 tasks

94.4%

composite

Outputs

The actual rendered output. Open any tile to view it in a popup, or compare the same task across every model.

claude-opus-4-8 Animation output - Simple UI animation

Compare modelsANIM-0001judge 3.3/5

claude-opus-4-8 Animation output - Complex animation scene

Compare modelsANIM-0002judge 4.3/5

claude-opus-4-8 Auth screen output - Register / login screen

Compare modelsAUTH-0001judge 3.8/5

claude-opus-4-8 UI components output - Keyboard-operable modal dialog with focus trap and return

Compare modelsAYGAT-0001100.0%

claude-opus-4-8 UI components output - Keyboard-operable select-only combobox with a listbox popup

Compare modelsAYGAT-00020.0%

claude-opus-4-8 UI components output - Keyboard-operable menu button with arrow-key navigation

Compare modelsAYGAT-000398.1%

claude-opus-4-8 UI components output - Keyboard-operable tabs with roving tabindex and arrow-key activation

Compare modelsAYGAT-000498.1%

claude-opus-4-8 UI components output - Keyboard-operable accordion with disclosure buttons and arrow navigation

Compare modelsAYGAT-0005100.0%

claude-opus-4-8 UI components output - Keyboard-operable ARIA radiogroup with roving tabindex

Compare modelsAYGAT-000698.1%

claude-opus-4-8 Dashboard output - SaaS analytics dashboard

Compare modelsDASH-0001judge 4.3/5

claude-opus-4-8 SVG and graphics output - Bar chart with exact proportional heights for [4, 10, 6, 8, 12]

Compare modelsDIAGR-000125.0%

claude-opus-4-8 SVG and graphics output - Four-node pipeline flow graph with exactly four edges

Compare modelsDIAGR-0002100.0%

claude-opus-4-8 SVG and graphics output - Horizontal stacked bar with proportional segments and a legend

Compare modelsDIAGR-0003100.0%

claude-opus-4-8 SVG and graphics output - Org chart with three reporting edges and a fixed hierarchy

Compare modelsDIAGR-0004100.0%

claude-opus-4-8 SVG and graphics output - Gantt chart with proportional start offsets and durations

Compare modelsDIAGR-0005100.0%

claude-opus-4-8 SVG and graphics output - Grouped two-series bar chart across three categories with a legend

Compare modelsDIAGR-0006100.0%

claude-opus-4-8 SVG and graphics output - Five-node hub-and-satellite network with exactly six edges

Compare modelsDIAGR-0007100.0%

claude-opus-4-8 Marketing page output - Marketing landing page

Compare modelsMKT-0001judge 3.8/5

claude-opus-4-8 UI components output - Stat card built to a frozen token sheet

Compare modelsPIXEL-0001100.0%

claude-opus-4-8 UI components output - Three overlapping layers in an exact z-order and position

Compare modelsPIXEL-0002100.0%

claude-opus-4-8 UI components output - Flex toolbar with an exact gap and exact button geometry

Compare modelsPIXEL-0003100.0%

claude-opus-4-8 UI components output - Two-column grid split with exact column widths and gutter

Compare modelsPIXEL-0004100.0%

claude-opus-4-8 UI components output - Pill CTA button built to an exact box and token sheet

Compare modelsPIXEL-0005100.0%

claude-opus-4-8 UI components output - Progress bar with an exact fill width and track tokens

Compare modelsPIXEL-0006100.0%

claude-opus-4-8 UI components output - Segmented pill nav with an exact active pill and gap

Compare modelsPIXEL-0007100.0%

Compare modelsPIXEL-0008100.0%

claude-opus-4-8 Websites output - Header nav that collapses to a menu button below a breakpoint

Compare modelsRESPO-00010.0%

claude-opus-4-8 Websites output - Feature grid that reflows from one to two to three columns

Compare modelsRESPO-00020.0%

claude-opus-4-8 Websites output - Sidebar and content layout that never overflows, including at the tablet width

Compare modelsRESPO-0003100.0%

claude-opus-4-8 Websites output - Responsive nav that both collapses at the breakpoint and toggles open on mobile

Compare modelsRESPO-0004100.0%

claude-opus-4-8 Websites output - Gallery grid that widens from one to four columns as the viewport grows

Compare modelsRESPO-00050.0%

claude-opus-4-8 Websites output - Headline type scale that steps up at the tablet and desktop breakpoints

Compare modelsRESPO-0006100.0%

claude-opus-4-8 Websites output - CTA buttons that stack full-width on mobile and sit in a row on desktop

Compare modelsRESPO-00070.0%

claude-opus-4-8 svg-scene output - Sydney Harbour at golden hour (SVG scene)

Compare modelsSCENE-0001judge 3.5/5

claude-opus-4-8 Websites output - Reproduce a three-tier SaaS pricing page from its screenshot

Compare modelsSCREE-000197.8%

claude-opus-4-8 Websites output - Reproduce a dark analytics dashboard overview from its screenshot

Compare modelsSCREE-00020.0%

claude-opus-4-8 Websites output - Reproduce an editorial blog article page from its screenshot

Compare modelsSCREE-000391.7%

claude-opus-4-8 Websites output - Reproduce a split product-feature section from its screenshot

Compare modelsSCREE-00040.0%

claude-opus-4-8 Websites output - Reproduce the Banja home page hero from its screenshot

Compare modelsSCREE-000592.8%

claude-opus-4-8 SVG and graphics output - Single-concept flat icon (coffee cup with steam)

Compare modelsSVG-0001100.0%

claude-opus-4-8 SVG and graphics output - Simple house icon (square, triangle roof, door)

Compare modelsSVG-0002100.0%

claude-opus-4-8 SVG and graphics output - Flat multi-object scene (sun, hills, hot-air balloon)

Compare modelsSVG-0003100.0%

Compare modelsSVG-0004100.0%

claude-opus-4-8 SVG and graphics output - Tiny bar chart for values [3, 7, 5, 9]

Compare modelsSVG-0005100.0%

claude-opus-4-8 SVG and graphics output - Traffic light with only the green light lit

Compare modelsSVG-0006100.0%

claude-opus-4-8 UI components output - Accessible pricing card with a monthly/annual toggle

Compare modelsUI-0001100.0%

claude-opus-4-8 UI components output - Accessible accordion FAQ that expands on click

Compare modelsUI-0002100.0%

claude-opus-4-8 UI components output - Tabbed interface that switches panels on click

Compare modelsUI-0003100.0%

claude-opus-4-8 UI components output - Signup form with inline email validation

Compare modelsUI-0004100.0%

claude-opus-4-8 UI components output - Stat and testimonial card grid

Compare modelsUI-0005100.0%

claude-opus-4-8 Websites output - Landing page with a working mobile nav toggle

Compare modelsWEB-000124.0%

claude-opus-4-8 Websites output - Dashboard with a sortable data table

Compare modelsWEB-0002100.0%

claude-opus-4-8 Websites output - Pricing page with a working billing toggle and FAQ accordion

Compare modelsWEB-0003100.0%

Objective tasks

Programming, Australian legal and accounting, graded by execution. 27 of 34 scored a perfect 100.0%; the rest are below. Open the answer in a popup, or compare it across every model.

Task	Domain	Difficulty	Objective	pass@1	Output
AUSFA-0011	Australian accounting	hard	66.7%	0.0%	Compare Open
AUSFA-0012	Australian accounting	hard	33.3%	0.0%	Compare Open
AUSFA-0013	Australian accounting	hard	66.7%	0.0%	Compare Open
AUSFA-0014	Australian accounting	hard	0.0%	0.0%	Compare Open
CODE-0004	Programming	easy	0.0%	0.0%	Compare Open
CODE-0006	Programming	hard	0.0%	0.0%	Compare Open
LAW-0003	Australian law	hard	0.0%	0.0%	Compare Open