Compare LLM Endpoints by Benchmark

What you can inspect

Inspectable by default: public runs, raw outputs, prompt details, score, latency, sample counts, and methodology.

• Public runs: Representative benchmark runs are linked and inspectable.
• Raw outputs: Review model responses instead of relying only on scores.
• Prompt details: See prompt mode, setup, and run configuration.
• Score + latency: Compare quality and speed in the same view.
• Sample counts: Check how much data supports each result.
• Methodology: Understand comparability, caveats, and scoring.

Benchmark Results

Each benchmark page groups comparable public runs for one task suite, anchored to a fixed prompt, scoring method, and sample size. MMLU and MATH have the most coverage.

Model Families and Endpoints

A model family is the underlying model identity. A hosted endpoint is a specific provider's deployment of that model. The same model family can produce different benchmark results depending on which provider hosts it.

Runs

Public runs record the endpoint, benchmark, prompt mode, lifecycle state, score, latency, sample count, and raw per-example outputs. Use canonical-prompt runs for the fairest cross-provider comparisons.

Methodology

Benchscope records full evaluation metadata at every step. Read the methodology to understand how runs are defined, what makes results comparable, and what caveats apply.

Flagship Benchmarks

• MMLU — 57-subject knowledge breadth across STEM, humanities, and professional domains
• MATH — competition-level mathematical problem solving across algebra, geometry, and number theory
• GSM8K — grade school math word problems requiring multi-step arithmetic reasoning
• IFEval — instruction-following compliance verified by programmatic rule checks
• MuSR — multi-step compositional reasoning over narrative contexts
• All benchmarks on Benchscope

Provider Benchmark Results

• Groq Benchmark Results
• Together AI Benchmark Results
• Cerebras Benchmark Results
• Model Families and Hosted Endpoints
• How results are defined and compared

Featured Model Pages

• Llama 3.3 70B benchmark results — Compare Llama 3.3 70B benchmark results across providers and endpoints.
• Qwen 3 235B A22B benchmark results — Compare Qwen 3 235B A22B benchmark results across providers and endpoints.
• DeepSeek R1 benchmark results — Review DeepSeek R1 benchmark scores, latency, and provider coverage.
• Llama 4 Scout 17B benchmark results — Review Llama 4 Scout 17B benchmark scores, latency, and provider coverage.
• GPT-OSS 120B benchmark results — Review GPT-OSS 120B benchmark scores, latency, and provider coverage.

Editorial Comparisons

• Best LLM Endpoint for MMLU — ranked by 57-subject knowledge score
• Best LLM Endpoint for MATH — ranked by competition-level math score
• Best LLM Endpoint for GSM8K — ranked by grade school math reasoning score
• Llama 3.3 70B on Groq vs Together AI — same model, different providers