Independent first
Third-party evaluator results are separated from provider launch claims.
EVALUATION INTELLIGENCE
AI benchmarks.
Configuration attached.
Benchmark scores are evidence, not universal truth. SXF stores who ran the evaluation, which version, which model configuration and which results are actually comparable.
12benchmark definitions
30independent observations
4vendor-reported
2026-10-04verified
INTERACTIVE EXPLORERCost vs score. Context vs score. Comparable groups only.Analyze the current evidence without collapsing benchmarks into one universal ranking.
Open Explorer ↗
COMPARABILITY RULE
Same name does not mean same measurement.
SXF only treats results as numerically comparable when benchmark identity, version, evaluator methodology and comparable group line up. Tool use, fallback behavior and reasoning effort remain part of the observation.
Version pinned
Live leaderboards can change methodology. Every snapshot stores its observed date and benchmark version.
Configuration pinned
Reasoning effort, tools and fallback behavior stay attached to the score.
No synthetic IQ
Different benchmarks are not collapsed into a homemade universal intelligence score.
BENCHMARK REGISTRY
What SXF tracks.
independentactive
Artificial Analysis Intelligence Index 4.3.2
composite intelligence · score · higher-is-better
6 observations Open benchmark evidence ↗independentactive
AA-Briefcase 1.1
agentic knowledge work · elo · higher-is-better
3 observations Open benchmark evidence ↗independentactive
GDPval-AA 2.1
professional knowledge work · elo · higher-is-better
3 observations Open benchmark evidence ↗independentactive
AutomationBench-AA
business automation · task-success · higher-is-better
3 observations Open benchmark evidence ↗independentactive
Terminal-Bench 4.0
agentic coding · task-success · higher-is-better
3 observations Open benchmark evidence ↗independentunder-review
SciCode
scientific coding · score · higher-is-better
3 observations Open benchmark evidence ↗independentactive
Humanity's Last Exam
multidisciplinary reasoning · accuracy · higher-is-better
3 observations Open benchmark evidence ↗independentunder-review
CritPt
frontier science reasoning · score · higher-is-better
3 observations Open benchmark evidence ↗independentactive
AA-LCR 1.1
long context reasoning · score · higher-is-better
3 observations Open benchmark evidence ↗vendor-reportedactive
AutomationBench
professional workflows · task-success · higher-is-better
1 observations Open benchmark evidence ↗vendor-reportedactive
Terminal-Bench 4.0
agentic coding · task-success · higher-is-better
1 observations Open benchmark evidence ↗vendor-reportedactive
Humanity's Last Exam
multidisciplinary reasoning · accuracy · higher-is-better
2 observations Open benchmark evidence ↗MODEL COVERAGE