Claude Haiku 5.5 came out on . It answered the same 24 prompts as every other model on this site, word for word, at low and at max effort. It scored 94 at low effort and 97 at max effort, out of 100. Ranked 1st of 8 at max effort, level with Fable 5.1, Fable 5 and Opus 5.5.
Against Haiku 4.5
Haiku 5.5 scored 97 at max effort and Haiku 4.5 72 at its one setting, 25.5 points apart.
Scored higher by 5 points or more: Haiku 5.5 on building web pages, 98 to 63; Haiku 5.5 on code, 100 to 76; Haiku 5.5 on planning and architecture, 91 to 77; Haiku 5.5 on everyday, 100 to 71.
Haiku 5.5’s median max effort answer took 1 min 56 s and cost 1.5¢ at API prices, and Haiku 4.5’s median answer took 6.8 s and cost 0.5¢. A median Haiku 5.5 answer cost 3.3 times as much.
Of the 24 prompts, Haiku 5.5 passed more checks on 16, Haiku 4.5 on none, and the two passed the same number on 8. The full comparison.
Cost and speed
Haiku 5.5’s median max effort answer cost 1.5¢ at API prices, the second lowest of the eight models at their highest setting, after Haiku 4.5.
Its median max effort answer took 1 min 56 s; three models were quicker at their highest setting.
Costs are what the same tokens would cost at Anthropic’s API prices on the day it was tested, not what anyone paid.
Where it missed
At max effort it passed every check on 20 of the 24 prompts. These are the 4 checks it missed at max effort:
- B03 A kanban board that survives a reload: A card can be moved with the arrow keys and keeps focus
- A01 Plan a booking app for a dog groomer: The first phase is small and usable before payments exist
- A03 Rename a column with no downtime: Drops the old column last, after a waiting period
- A04 Turn a messy feature brief into a spec: Asks about an invitee already in another team and about pending invites on downgrade
What max effort changed
Max passed 5 more checks, cost 10.2 times as much and took 10.6 times as long. Max effort scored higher on 5 prompts, lower on 2 and the same on 17. What max effort changed for every model.
Every answer
Every answer from Haiku 5.5 is on its model page, next to its checks, time, tokens and cost. Compare it with any other model.