Price per job versus rubric score, nine models, on a logarithmic price axis.
A scatter plot. The horizontal axis is price per job on a logarithmic scale from about ten cents to ten dollars; the vertical axis is rubric score from zero to twenty. Points cluster near the top of the score axis across a wide spread of prices, forming a flat capability ceiling. A faint teal band marks the seventeen-to-twenty score region with a dashed teal guide along its floor. Three verified points share the score of twenty: deepseek-v4-pro at eighteen cents (open-weight), gpt-5.5 at one dollar thirty-two, and claude-fable-5 at nine dollars seventy-four, a roughly fifty-four times price spread for the identical grade. Six approximate run-level points are drawn faintly without labels. Marker shape encodes vendor: circle for Anthropic, square for OpenAI, triangle for Google, diamond for other.
20
15
10
5
0
$0.10
$1.00
$10.00
capability ceiling (17–20)
Approximate run-level: gemini-3.5-flash (google), about $1.52 per job, score 19 of 20.
Approximate run-level: gpt-5.4 (openai), about $1.24 per job, score 16 of 20.
Approximate run-level: gemini-3.1-pro (google), about $0.99 per job, score 18 of 20.
Approximate run-level: claude-haiku-4.5 (anthropic), about $0.78 per job, score 13 of 20.
Approximate run-level: grok-4.3 (other), about $0.42 per job, score 14 of 20.
Approximate run-level: deepseek-v3.2 (other), about $0.29 per job, score 13 of 20.
same 20/20, ~54× the price
Verified: deepseek-v4-pro (other, open-weight), $0.18 per job, score 20 of 20.
Verified: gpt-5.5 (openai), $1.32 per job, score 20 of 20.
Verified: claude-fable-5 (anthropic), $9.74 per job, score 20 of 20.
deepseek-v4-pro
$0.18
open-weight
claude-fable-5
$9.74
gpt-5.5
$1.32
price per job (log scale, sum of 3 graded tasks)
score (out of 20)
Four models tie at a perfect 20/20, yet the price of that identical grade
runs from $0.18 (deepseek-v4-pro, open-weight) to $9.74 (claude-fable-5)
— a flat ceiling over roughly 54× the cost.
verified (exact, labeled): deepseek-v4-pro $0.18 · gpt-5.5 $1.32 · claude-fable-5 $9.74 · all 20/20
6 approximate run-level points (shape only) · shape = vendor: circle anthropic / square openai / triangle google / diamond other