Table
Ranks only the selected comparable group by the benchmark's declared metric direction.
EVALUATION EXPLORER V1
Benchmark evidence.
Decision views, not one score.
Filter by benchmark, evaluator, provider and comparable group. Switch between ranked evidence, cost-vs-score and context-vs-score views without mixing incompatible runs.
12benchmarks
34observations
6models evaluated
2026-10-04verified
INTERACTIVE ANALYSIS
Choose one measurement frame.
The explorer only ranks observations inside the selected benchmark and comparable group. Missing coverage stays missing; under-review benchmarks carry a visible warning.
| Model | Score | Context | Workload cost | Reasoning | Evidence |
|---|---|---|---|---|---|
| Claude Opus 5.5Anthropic | 58 | 1,000,000tokens | $0.6100K input + 10K output | max | Evidence ↗ |
| Claude Fable 5.1Anthropic | 53 | 1,000,000tokens | $1.5100K input + 10K output | max | Evidence ↗ |
| GPT-6 AstraOpenAI | 51 | 1,050,000tokens | $1.5100K input + 10K output | high | Evidence ↗ |
| Grok 4.7xAI | 46 | 500,000tokens | $0.26100K input + 10K output | xhigh | Evidence ↗ |
| GPT-6 SolOpenAI | 42 | 1,050,000tokens | $0.3100K input + 10K output | high | Evidence ↗ |
| Gemini 3.8 FlashGoogle | 41 | 1,048,576tokens | $0.1125100K input + 10K output | high | Evidence ↗ |
READING THE VIEWS
Pareto means efficient, not universally best.
A model is Pareto-efficient when no other visible model is both better on the selected benchmark and better on the second axis. Cost uses the SXF Standard workload assumption of 100K uncached input + 10K output tokens; context uses the provider-published context window.
Cost vs score
Highlights models that offer a non-dominated tradeoff between direct token cost and benchmark score.
Context vs score
Shows whether a larger context window trades off against the selected evaluation result.
No speed view yet
Latency and throughput stay out until SXF has a methodology-pinned speed dataset with matching model configurations.