AskClash model intelligence

AI Model Benchmarks

Compare verified AI work, intelligence, reasoning effort, open-weight models, and defined API workload cost with clear, shareable charts.

AI benchmark charts worth sharing

The interactive library turns the current AskClash model snapshot into practical comparisons for choosing a model by capability, cost, and workflow.

Compare measured cohort-relative benchmark scores across all, open-weight, or proprietary models; editorially floored rows are excluded.

Compare model scores using the disclosed cost of exactly 1M input and 200K output tokens.

Compare verified engineering pass rates with actual task cost, output tokens, agent steps, and completion time.

Compare all published models, or focus on one, to see how reasoning effort changes verified work and task cost.

Compare GPT-5.6 Sol, Terra, and Luna with Claude, Grok, Gemini, and GPT-5.5 using one defined workload.

Compare models that publish both DeepSWE and Terminal-Bench, with SWE-bench retained as separate context.

Compare GPQA, HLE, and AAII only for models with complete published coverage.

Compare MMMU-Pro, CharXiv, and OSWorld only when all three scores are published.

Find high-scoring models below $5 for 1M input plus 200K output tokens, excluding zero-price placeholders.