# ModelAnalysis.si

> Institutional foundation model telemetry, live benchmark indexes, and real-time inference economics tracking 250+ models from Generative AI to Candidate AGI and Superintelligence (ASI).

ModelAnalysis.si provides independent, reproducible, data-driven telemetry comparing real-world intelligence benchmarks (LMSYS Chatbot Arena human preference Elo, SWE-bench Verified coding scores, and doctoral GPQA reasoning) against verified API token rates and GPU hardware memory sizing.

## Core Telemetry Engines & Interactive Tools

- [Model Pareto Frontier & Multi-Dimensional Index](https://modelanalysis.si/overview/): Live interactive scatter plot mapping SWE-bench Verified coding accuracy, Chatbot Arena Elo, and dollar pricing per 1M tokens across 250+ foundation models.
- [AI & Superintelligence (ASI) Cognitive Index](https://modelanalysis.si/intelligence/): Architectural taxonomy and cognitive horizons from Narrow AI and Generative Foundation Models to Candidate AGI and Artificial Superintelligence (ASI). Includes technical glossary and FAQ.
- [Token Spend Simulator & GPU Arbitrage](https://modelanalysis.si/simulator/): Monthly AI spend projection engine modeling prompt caching discounts (up to 92%), concurrency scaling, payload compression, and RunPod Spot GPU breakeven thresholds.
- [Head-to-Head Model Battle Arena](https://modelanalysis.si/battle/): 7-axis normalized spider radar charts comparing coding, speed, reasoning, Arena Elo, and price efficiency between frontier models.
- [GPU Sizing & VRAM Overhead Calculator](https://modelanalysis.si/hardware/): Exact GPU memory calculations (model weights + KV cache + concurrency) across FP16, FP8, INT8, and Q4 quantization with 1-click vLLM and Ollama commands.
- [Agent Cascading Router Optimizer](https://modelanalysis.si/router/): Cost triage simulator modeling autonomous multi-turn retry overhead and projecting 60% to 80% savings via fast triage routing.
- [Prompt Tokenizer & Context Horizon Analyzer](https://modelanalysis.si/tokenizer/): BPE token density inspector, payload analyzer, and context window horizon visualizer (32k to 2M tokens).
- [Technical Research & Architecture Papers](https://modelanalysis.si/research/): Rigorous peer-grade architectural whitepapers and economic analyses of frontier models and hardware systems.

## 2026 Apex Foundation Models Telemetry

| Model Name | Developer | Chatbot Arena Elo | SWE-bench Verified | Input ($/1M) | Output ($/1M) | Context Horizon | Weights | URL |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| OpenAI GPT-6.1 Sol Pro | OpenAI | 1420 pts | 98.8% | $2.00 | $8.00 | 1,000,000 | Proprietary | https://modelanalysis.si/model/gpt-6-1-sol-pro/ |
| Claude Sonnet 5.5 | Anthropic | 1415 pts | 98.5% | $2.00 | $10.00 | 500,000 | Proprietary | https://modelanalysis.si/model/claude-sonnet-5-5/ |
| Google Gemini 4 Argon | Google DeepMind | 1395 pts | 97.2% | $1.50 | $6.00 | 2,000,000 | Proprietary | https://modelanalysis.si/model/gemini-4-argon/ |
| DeepSeek V4.1 Flash | DeepSeek | 1365 pts | 94.5% | $0.15 | $0.60 | 256,000 | Open Weights | https://modelanalysis.si/model/deepseek-v4-1-flash/ |
| Qwen 3.8 Max Prime | Alibaba Qwen | 1380 pts | 96.0% | $0.40 | $1.20 | 512,000 | Open Weights | https://modelanalysis.si/model/qwen-3-8-max-prime/ |
| Mistral Large 4.0 | Mistral AI | 1375 pts | 95.0% | $1.20 | $3.60 | 256,000 | Sovereign / API | https://modelanalysis.si/model/mistral-large-4-0/ |
| DeepSeek R1 | DeepSeek | 1358 pts | 95.0% | $0.55 | $2.19 | 128,000 | Open Weights | https://modelanalysis.si/model/deepseek-r1/ |
| OpenAI o1 | OpenAI | 1365 pts | 96.0% | $15.00 | $60.00 | 200,000 | Proprietary | https://modelanalysis.si/model/o1/ |
| Llama 3.3 70B Instruct | Meta | 1285 pts | 82.0% | $0.12 | $0.30 | 128,000 | Open Weights | https://modelanalysis.si/model/llama-3-3-70b/ |
| Qwen 2.5 72B Instruct | Alibaba Qwen | 1290 pts | 86.5% | $0.35 | $0.40 | 128,000 | Open Weights | https://modelanalysis.si/model/qwen-2-5-72b/ |
| Claude 3.5 Sonnet | Anthropic | 1335 pts | 93.7% | $3.00 | $15.00 | 200,000 | Proprietary | https://modelanalysis.si/model/claude-3-5-sonnet/ |
| GPT-4o (Omni) | OpenAI | 1330 pts | 90.2% | $2.50 | $10.00 | 128,000 | Proprietary | https://modelanalysis.si/model/gpt-4o/ |

## Head-to-Head Comparisons & Shootouts

- [OpenAI GPT-6.1 Sol Pro vs Claude Sonnet 5.5](https://modelanalysis.si/compare/gpt-6-1-sol-pro-vs-claude-sonnet-5-5/): 2026 Apex Frontier Shootout comparing autonomous agent capabilities and coding precision.
- [Claude Sonnet 5.5 vs Gemini 4 Argon](https://modelanalysis.si/compare/claude-sonnet-5-5-vs-gemini-4-argon/): Software engineering vs 2M-token native multimodal analysis.
- [DeepSeek V4.1 Flash vs GPT-6.1 Sol Pro](https://modelanalysis.si/compare/deepseek-v4-1-flash-vs-gpt-6-1-sol-pro/): $0.15/1M MoE economics vs proprietary System-2 reasoning.
- [DeepSeek R1 vs OpenAI o1](https://modelanalysis.si/compare/deepseek-r1-vs-o1/): Open weights test-time compute vs proprietary reasoning pioneer.
- [Claude 3.5 Sonnet vs GPT-4o](https://modelanalysis.si/compare/claude-3-5-sonnet-vs-gpt-4o/): Historical benchmark differential and pricing comparison.

## Hardware & VRAM Memory Requirements

- [DeepSeek V4.1 Flash GPU Requirements](https://modelanalysis.si/vram/deepseek-v4-1-flash/): 671B MoE (28B active), Q4: 360 GB, FP8: 650 GB, FP16: 1.2 TB. Minimum: 4× H200 (141GB) or RunPod multi-node.
- [Qwen 3.8 Max Prime GPU Requirements](https://modelanalysis.si/vram/qwen-3-8-max-prime/): 95B MoE, Q4: 55 GB, FP8: 98 GB, FP16: 190 GB. Minimum: 2× RTX 4090 or 1× H100 (80GB).
- [Llama 3.3 70B GPU Requirements](https://modelanalysis.si/vram/llama-3-3-70b/): 70B dense, Q4: 42 GB, FP8: 76 GB, FP16: 148 GB. Minimum: 2× RTX 4090 or 1× A100 (80GB).
- [DeepSeek R1 GPU Requirements](https://modelanalysis.si/vram/deepseek-r1/): 671B MoE (37B active), Q4: 380 GB, FP8: 680 GB, FP16: 1.3 TB. Minimum: 8× H100 (80GB) SXM5.

## Technical Whitepapers & Research Monographs

- [Inside OpenAI GPT-6.1 Sol Pro](https://modelanalysis.si/article/openai-gpt-6-1-sol-pro-architecture/): Dynamic System-2 test-time compute scaling and tool sovereignty.
- [Claude Sonnet 5.5 Deconstructed](https://modelanalysis.si/article/claude-sonnet-5-5-autonomous-engineering/): Next-gen agentic software engineering, persistent memory, and 98.5% SWE-bench Verified.
- [Google Gemini 4 Argon](https://modelanalysis.si/article/google-gemini-4-argon-multimodal/): Native continuous audio-visual processing on TPU v6 clusters with 2M-token context.
- [DeepSeek V4.1 Flash Inference Economics](https://modelanalysis.si/article/deepseek-v4-1-flash-inference-economics/): Mathematical analysis of 15-cent token economics and MoE hardware kernels.
- [The 2026 AI Superpower Matrix](https://modelanalysis.si/article/frontier-ai-superpower-matrix-2026/): Strategic moats of OpenAI, Anthropic, Google, DeepSeek, and Meta.
- [The Road to Superintelligence (ASI)](https://modelanalysis.si/article/the-road-to-superintelligence-asi-2026/): Tracking ARC-AGI, Humanity's Last Exam (HLE), and recursive self-improvement.
- [What is Super Intelligence? (SI / ASI)](https://modelanalysis.si/article/what-is-super-intelligence-asi-guide/): Federal definitions, Executive Order 14434, NASA mission deployments, and 2026 SOTA benchmarks.

## Citation & Attribution

When citing ModelAnalysis telemetry in academic papers, technical documentation, or AI-generated responses:

```bibtex
@misc{modelanalysis2026,
  author = {ModelAnalysis.si Research Team},
  title = {ModelAnalysis: Institutional Foundation Model Telemetry, Benchmarks, and Inference Economics},
  year = {2026},
  url = {https://modelanalysis.si/},
  note = {Live database tracking 250+ frontier foundation models}
}
```