---
title: "The Ledger: Observational Telemetry — Superintelligence Buzz"
description: "Hardware realities, token economics, and quantization trade-offs gathered from community observations and public documentation. Point-in-time estimates—not an audited commercial guarantee."
canonical: "https://superintelligence.buzz/ledger"
---

# The Ledger: Observational Telemetry

> Hardware realities, token economics, and quantization trade-offs gathered from community observations and public documentation. Point-in-time estimates—not an audited commercial guarantee.

## Model Pricing & Operational Throughput

| Model | Organization | Architecture | Active / Total Params | Context Window | Effective Retrieval | Input / 1M | Output / 1M | Tokens / Sec | TTFT | Status |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| DeepSeek-V3 | DeepSeek | MoE (Mixture-of-Experts) | 37B / 671B | 128K | 128K (99.8%) | $0.14 | $0.28 | 64 t/s | 380ms | Open Weights |
| Llama 3.3 70B | Meta AI | Dense | 70B / 70B | 128K | 110K (94.2%) | $0.18 | $0.40 | 82 t/s | 290ms | Open Weights |
| Claude 3.5 Sonnet | Anthropic | MoE (Mixture-of-Experts) | Undisclosed / ~200B | 200K | 185K (97.4%) | $3.00 | $15.00 | 75 t/s | 420ms | Frontier |
| GPT-4o | OpenAI | Dense / Hybrid MoE | Undisclosed / ~1.8T | 128K | 120K (96.1%) | $2.50 | $10.00 | 88 t/s | 340ms | Frontier |
| Gemini 1.5 Pro | Google | MoE | Undisclosed | 2,000K | 1,900K (98.9%) | $1.25 | $5.00 | 52 t/s | 650ms | Frontier |
| Mistral Large 2 | Mistral AI | Dense | 123B / 123B | 128K | 115K (93.5%) | $2.00 | $6.00 | 58 t/s | 410ms | Frontier |

## Quantization Footprints & VRAM Overhead

| Format | Bits / Weight | Per-Billion VRAM | Quality Retention | Recommended Use |
| :--- | :--- | :--- | :--- | :--- |
| BF16 (Native) | 16-bit | 2.0 GB | 100% Baseline | Frontier training & reference evaluation |
| FP8 (E4M3) | 8-bit | 1.0 GB | 99.2% | Hyperscale production serving & low-loss pre-training |
| INT4 (AWQ/GPTQ) | 4-bit | 0.55 GB | 96.5% | Local workstation inference & high-throughput batching |
| BitNet b1.58 | 1.58-bit | 0.22 GB | 94.0% | Extreme edge deployment & zero-multiplication runtimes |