CYBERGIZER · MODEL INTELLIGENCE
Choose a model for your workload.
Compare hosted models and providers using published metadata. Estimate token spend, then validate quality on your own tasks.
Presets are editable workload assumptions, not measured model performance. Monthly estimate = requests × (input tokens × input rate + output tokens × output rate) ÷ 1,000,000. USD, token costs only; excludes images, audio, caching, retries, tools, hosting and taxes. Context check counts input + output tokens.
Quality signal and output cost
Plot a named, higher-is-better benchmark against provider output price. Matching historical evaluation rows or named user-imported metric rows are plotted. Historical results do not validate the currently hosted provider configuration. Source-reported latency is not a quality score.
Side-by-side shortlist
Select up to four model/provider rows to compare.
Historical evaluation provenance
Six exact model IDs have archived Open LLM Leaderboard MMLU-Pro raw accuracy (0–1), 5-shot task v0.1, from 2024–2025. Scores describe the evaluated revisions, precision and chat templates. Provider API deployments may use different weights, quantization or serving settings. These are historical reference signals, not current provider performance measurements or Cybergizer evaluations.
| Exact model ID | Raw accuracy | Evaluation | Precision / revision |
|---|---|---|---|
| meta-llama/Llama-3.1-8B-Instruct ↗ | 0.37982 | 2025-02-13 | float160e9e39f249a16976918f6564b8830bc894c89659 |
| meta-llama/Llama-3.3-70B-Instruct ↗ | 0.53316 | 2025-02-13 | bfloat16246a3e843c8cfea74524b24fd66934372e7b36b4 |
| Qwen/Qwen2.5-Coder-7B-Instruct ↗ | 0.33519 | 2025-02-13 | bfloat16f784f10a7b2aac91bd26e6dbe7dccce691cd4ac5 |
| Qwen/Qwen2.5-Coder-32B-Instruct ↗ | 0.44132 | 2025-02-13 | float16b47205940b83b5b484577359f71ee7b88472df67 |
| Qwen/Qwen2.5-72B-Instruct ↗ | 0.56258 | 2025-02-13 | bfloat16d3d951150c1e5848237cd6a7ad11df4836aee842 |
| NousResearch/Hermes-3-Llama-3.1-70B ↗ | 0.47266 | 2025-02-13 | bfloat16093242c69a91f8d9d5b8094c380b88772f9bd7f8 |
Raw accuracy is not the leaderboard’s normalized score. Comparisons should inspect the linked evaluation configuration. Download exact provenance JSON ↗
Sources, limitations and CSV format
Source: Hugging Face public model catalogue. Provider-specific pricing is USD per million tokens. The catalogue is not the entire model market. A model's image modality does not guarantee every provider's image support. Tool and structured-output flags are provider-specific. Probe latency and throughput are operational signals, not quality benchmarks or service guarantees. Unknown values stay unknown and sort last.
Import appends locally to this session. No inference calls, credentials or file uploads. Required CSV headers: model,provider. Optional: input_usd_per_million,output_usd_per_million,context_tokens,tools,vision,structured,latency_ms,throughput_tps,benchmark_name,benchmark_score,source_url,as_of. Boolean fields: true / false; empty = unknown. Benchmark scores must use one consistent, higher-is-better metric and test version. Select its exact name before sorting. Imported data is labelled unverified. Existing public catalogue rows are preserved.