88.8
Fixed Income Benchmark for Rigorous Evaluation
- Models
- 14
- Tasks
- 202
- Rubric criteria
- 1,754
- Updated
- Oct 8, 2026
Graded on the reasoning, not only the final number.
152 tasks are answered from a written brief and 50 are researched live from filings. Each has its own rubric of specific claims, and a panel of three judge models from different labs checks every answer against it. Cost and latency are shown alongside the score.
Leaderboard
#
06
haiku-5.5-high
Anthropic · high
Score
87.0
Question-answering
86.3
Agentic-research
89.1
Latency74.7s
Cost$0.010
08
deepseek-v4p1-flash
DeepSeek
Score
85.5
Question-answering
84.3
Agentic-research
89.3
Latency188s
Cost$0.024
11
gpt-5.6-terra
OpenAI · high
Score
83.7
Question-answering
84.8
Agentic-research
80.1
Latency88.8s
Cost$0.244
Score is the average across all tasks; the two track columns break it down. Latency is the median time per task, and cost is per task at list API prices. The Lab view shows each lab’s best model.
Trade-offs
Best trade-off · no model beats these on bothBetter on both — no model here yetBeaten on both by a model on the lineBest-by-Task · hypothetical, not a model