Public Calibration
Accounting Knowledge
Kaelum Bench · FMB-KNOW-ACC-01
This calibration view covers the concrete 60-task suite only. Its provisional 600-point hybrid score combines deterministic and rubric components, applied modifiers, and an independent AI judge. It exposes only sanitized score, execution, token, cost, and judge-identity aggregates. It contains no prompts, responses, expected solutions, rubrics, task IDs, provider generation IDs, keys, rationales, digests, or private file paths.
Accounting Knowledge · 60-task sub-benchmark
Calibration results
Sanitized lifecycle, coverage, score, token, and cost aggregates for FMB-KNOW-ACC-01. Results are provisional and specific to the recorded model configuration.
- Sub-benchmark
- Accounting Knowledge · FMB-KNOW-ACC-01
- Models
- 9 calibration runs
- Coverage
- 60 tasks per model
- Deterministic
- 487 deterministic points
- AI rubric
- 113 AI-judged rubric points
Execution detail
Model runs
Provisional hybrid results combine deterministic scoring, an independent AI judge, and applied modifiers.
Provisional · Configuration-specific calibration · Single runs; no statistical ranking.
OpenRouter (Wafer, authenticated recovery) · xhigh reasoning
Z.ai: GLM 5.2 — open run evidence
z-ai/glm-5.2
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
96.0%
576.1 / 600
463.1 / 487 deterministic
113.0 / 113 AI rubric
0.0 modifier adjustment
60 valid answers
0 invalid JSON responses
60 judge assessments
0 judge retries
171,630 total tokens
$1.15 total cost
OpenRouter (xai) · high reasoning
xAI: Grok 4.5 — open run evidence
x-ai/grok-4.5
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
95.9%
575.5 / 600
464.5 / 487 deterministic
111.0 / 113 AI rubric
0.0 modifier adjustment
60 valid answers
0 invalid JSON responses
60 judge assessments
58 judge retries
146,307 total tokens
$1.08 total cost
OpenRouter (morph) · reasoning-max-supported reasoning
MiniMax: MiniMax M3 — open run evidence
minimax/minimax-m3
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
80.5%
483.3 / 600
381.6 / 487 deterministic
108.0 / 113 AI rubric
-6.3 modifier adjustment
60 valid answers
0 invalid JSON responses
60 judge assessments
27 judge retries
131,370 total tokens
$1.03 total cost
OpenRouter (google-vertex) · high reasoning
Gemini 3.5 Flash — open run evidence
google/gemini-3.5-flash
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
90.3%
541.6 / 600
432.6 / 487 deterministic
109.0 / 113 AI rubric
0.0 modifier adjustment
58 valid answers
2 invalid JSON responses
58 judge assessments
0 judge retries
197,580 total tokens
$1.63 total cost
OpenRouter (azure) · max reasoning
Claude Sonnet 5 (Max) — open run evidence
anthropic/claude-sonnet-5
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
93.2%
558.9 / 600
445.9 / 487 deterministic
113.0 / 113 AI rubric
0.0 modifier adjustment
60 valid answers
0 invalid JSON responses
60 judge assessments
0 judge retries
277,544 total tokens
$2.16 total cost
OpenRouter (alibaba) · xhigh reasoning
DeepSeek V4 Flash — open run evidence
deepseek/deepseek-v4-flash
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
95.0%
569.9 / 600
457.9 / 487 deterministic
112.0 / 113 AI rubric
0.0 modifier adjustment
60 valid answers
0 invalid JSON responses
60 judge assessments
0 judge retries
208,196 total tokens
$0.9077 total cost
OpenRouter (together) · high reasoning
NVIDIA Nemotron 3 Ultra — open run evidence
nvidia/nemotron-3-ultra-550b-a55b
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
85.1%
510.5 / 600
407.5 / 487 deterministic
108.0 / 113 AI rubric
-5.0 modifier adjustment
60 valid answers
0 invalid JSON responses
60 judge assessments
0 judge retries
123,569 total tokens
$0.9641 total cost
OpenRouter (akashml/fp8) · reasoning-max-supported reasoning
Qwen3.6-35B-A3B — open run evidence
qwen/qwen3.6-35b-a3b
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
93.3%
559.5 / 600
449.5 / 487 deterministic
110.0 / 113 AI rubric
0.0 modifier adjustment
60 valid answers
0 invalid JSON responses
60 judge assessments
0 judge retries
309,508 total tokens
$1.09 total cost
OpenRouter (modelrun/fp4) · reasoning-max-supported reasoning
Kimi K2.6 — open run evidence
moonshotai/kimi-k2.6
Judge openai/gpt-5.6-sol · OpenRouter (Azure)
60 / 60
94.7%
567.9 / 600
454.9 / 487 deterministic
113.0 / 113 AI rubric
0.0 modifier adjustment
60 valid answers
0 invalid JSON responses
60 judge assessments
0 judge retries
346,974 total tokens
$1.72 total cost