5-Lens Empirical Quadrants, Pareto Frontier & 17-Facet AI Catalog
Replaces subjective analyst pay-to-play quadrants with reproducible telemetry across LLMs, CPU/Edge SLMs, 3D/Multimodal, Time-Series Foundation Models (Chronos-Bolt, TimesFM 2.0, Moirai-MoE, TimeGPT, PatchTST, Toto), Agent SDKs/Orchestrators, 7-Stage RAG, DSPy, Embeddings, MCP + A2A, Security, Tracing vs Evaluation, Memory Policies vs Vector/SQL Substrates, and Durable Execution.
All 17 Facets (16) agent_development_orchestration ai_security_and_guardrails automation_ipaas_dag_durable embeddings_and_rerankers evaluation_and_regression_testing llm_frontier_cloud llm_open_weights_cloud memory_behavior_policies multimodal_vision_audio_3d observability_tracing_monitoring programmatic_prompt_workflow_optimization protocols_mcp_and_a2a rag_context_pipeline slm_cpu_edge_quantized teleon_team_simulation_provenance timeseries_foundation_models vector_and_transactional_storage
5 Orthogonal Ranking Lenses
- Lens 1 (Empirical 2x2 Quadrant): X = Verified Execution Reliability vs Y = Architectural Composability and Governance Fit.
- Lens 2 (Constraint-Weighted Wave): Custom workload weights plus Tversky hard vetoes (
spdx_license, blast class, CPU-only). - Lens 3 (Exact Pareto Frontier): Non-dominated set across Quality, Latency, Cost, Cliff Retention, and Lock-in Freedom (orange nodes).
- Lens 4 (Jagged-Frontier Cliff Index): Perturbed neighbor retention ratio.
- Lens 5 (Team Marginal Value / WAR): Paired holdout lift net of coordination token tax.
| Component & Publisher | Primary Facet | Quadrant & Frontier | Lens Scores | Cliff & WAR | License |
|---|---|---|---|---|---|
| llama.cpp (GGML / AVX-512 / NEON) ggml-org ( inference_runtime) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER PARETO | Exec: 96, Fit: 94, Wave: 93.84 | Cliff: 93%, WAR: +0.26 | MIT |
| Intel OpenVINO GenAI Intel ( inference_runtime) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 94, Fit: 94, Wave: 92.08 | Cliff: 93%, WAR: +0.21 | Apache-2.0 |
| ONNX Runtime GenAI Microsoft ( inference_runtime) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 94, Fit: 93, Wave: 92.04 | Cliff: 91%, WAR: +0.21 | MIT |
| PyTorch ExecuTorch Meta ( inference_runtime) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 93, Fit: 93, Wave: 91.48 | Cliff: 91%, WAR: +0.2 | BSD-3-Clause |
| Mistral Small 3.1 24B (Q4) Mistral AI ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 91, Fit: 91, Wave: 91.06 | Cliff: 86%, WAR: +0.22 | Apache-2.0 |
| Qwen3-8B (GGUF Q4_K_M) Alibaba Qwen ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 93, Fit: 92, Wave: 90.9 | Cliff: 86%, WAR: +0.23 | Apache-2.0 |
| MLC WebLLM (WebGPU / Wasm) MLC-AI ( inference_runtime) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 91, Fit: 92, Wave: 89.68 | Cliff: 89%, WAR: +0.19 | Apache-2.0 |
| Qwen3-Coder-7B-Instruct (GGUF) Alibaba Qwen ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 92, Fit: 91, Wave: 89.66 | Cliff: 84%, WAR: +0.21 | Apache-2.0 |
| IBM Granite 3.2 8B Instruct (Q4) IBM ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 93, Fit: 92, Wave: 89.62 | Cliff: 86%, WAR: +0.2 | Apache-2.0 |
| Qwen3-Coder-3B-Instruct (Q4_K_M) Alibaba Qwen ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 94, Fit: 91, Wave: 88.86 | Cliff: 84%, WAR: +0.19 | Apache-2.0 |
| Phi-4-mini-instruct (Q4_K_M) Microsoft ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 93, Fit: 91, Wave: 88.62 | Cliff: 84%, WAR: +0.19 | MIT |
| DeepSeek-R1-Distill-Qwen-7B (Q4) DeepSeek ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 90, Fit: 90, Wave: 88.5 | Cliff: 82%, WAR: +0.2 | MIT |
| SmolLM2-1.7B-Instruct Hugging Face ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 93, Fit: 91, Wave: 86.02 | Cliff: 82%, WAR: +0.15 | Apache-2.0 |
| Google Gemma 3 4B IT (Q4) Google ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 92, Fit: 91, Wave: 85.78 | Cliff: 84%, WAR: +0.18 | Gemma |
| Meta Llama 3.2 3B Instruct (Q4) Meta ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 94, Fit: 83, Wave: 84.94 | Cliff: 82%, WAR: +0.16 | Llama-3.2 |
| BitNet b1.58 2B4T Microsoft Research ( model_family_version) |
slm_cpu_edge_quantized |
EMPIRICAL_LEADER | Exec: 91, Fit: 82, Wave: 84.86 | Cliff: 80%, WAR: +0.14 | MIT |