FIBRE

haiku-4.5

Anthropic · claude-haiku-4-5-20251001

Ranked 5 of 5 overall on 202 tasks. Strongest on monitoring at 84.8, weakest on credit analysis at 58.5.

Score
65.2
rank 5 of 5
Latency
41.8s
median per task
Cost
$0.085
per task at list prices

Capability profile

Mean score per category on question-answering, 0–100.
CreditMonitorStructuredResearchDiscovery

Score by category

5 categories, 152 tasks on question-answering.
Credit Analysis58.5 · 62
Monitoring84.8 · 20
Structured Products69.4 · 18
Research Intelligence74.8 · 10
Discovery & Prediction79.7 · 2

Where it sits

haiku-4.5 against the rest of the board. Everything else is drawn back so the one model reads.
No model here.Better on both axes at once.$0.01$0.02$0.05$0.1$0.2$0.5$165758595Cost per task · better ←Score · better ↑Best per taskhaiku-4.5
haiku-4.5 beaten on both by gpt-5.6-luna and 1 otherFrontier · 3 models nothing beats on both

Best per task is the diamond: for each of the 200 tasks every model attempted, the highest score any of them got, averaged — 95.0 against 91.3 for opus-5 used on everything. Nobody can build it. The choice is made with the result in hand, which a router does not have when it has to choose, so it marks the edge of what having 5 different models is worth rather than a score anyone can reach. It is also flattered by noise — the highest of 5 measurements rises with the count even when the models behind them do not improve — so it will drift up as models are added and is not comparable between boards with different numbers of them. 2 tasks are left out of it, for not having been attempted by every model. Its cost is measured the way the board measures a model’s, on the question-answering track that holds more of the tasks.

How it works when given tools

Behaviour across 150 agentic-research runs.
Delivered an answer
100%
of runs produced a deliverable
Mean steps
25.9
tool calls per run
Tools used per run
fetch_url 19.4sec_search 3.9web_search 1.6python_exec 1.0
How runs ended
max steps 67%final answer 33%

Where it lost the most

The lowest-scoring tasks on question-answering, with the rubric gates that fired where they are known.
FAIL
Spread Analysis
Credit Analysis · spread, discount margin, floating rate
penalty: Does not report the quoted margin of 125bp as the bond's current spread. The quoted margi…
-11.1
PARTIAL
Spread Analysis
Credit Analysis · spread, accrued interest, dirty price
5.3
PARTIAL
Total Return Projection
Credit Analysis · total return, currency hedge, FX
penalty: Does not conclude the EUR bond is unattractive based solely on the naive hedged yield com…
6.7
PARTIAL
Breakeven Analysis
Credit Analysis · breakeven, carry, roll down
9.7
PARTIAL
Curve Trade Analysis
Credit Analysis · rates, curve trade, butterfly
14.7