Engineering Leverage & Prompt-Cache Telemetry
Breakdown
| # | Model | Cost | Share | Tokens | Cache Hit % | CR/CW Reuse |
|---|---|---|---|---|---|---|
| 1 |
claude-opus-5-5
Claude Code
|
$257.54 | 84.8% | 459M | 95.3% | 20.2x |
| 2 |
gpt-6.1-sol
OpenAI Codex
|
$29.99 | 9.9% | 150M | 96.9% | 31.5x |
| 3 |
claude-opus-4-8
Claude Code
|
$7.15 | 2.4% | 4.67M | 94.1% | 18.4x |
| 4 |
claude-sonnet-5-5
Claude Code
|
$6.20 | 2.0% | 22M | 95.0% | 24.8x |
| 5 |
antigravity-model-1018
Antigravity
|
$116.85 | 38.8% | 1.03B | 94.2% | 42.3x |
| 6 |
gemini-3.8-flash-e
Antigravity
|
$58.98 | 19.6% | 520M | 93.7% | 36.8x |
| 7 |
gemini-4-argon
Antigravity
|
$34.24 | 8.4% | 180M | 94.0% | 31.2x |
| 8 |
antigravity-model-1029
Antigravity
|
$22.90 | 7.6% | 202M | 93.5% | 29.6x |
| 9 |
antigravity-model-1417
Antigravity
|
$11.45 | 3.8% | 101M | 93.4% | 25.1x |
| 10 |
gpt-6-luna
OpenAI Codex
|
$0.14 | 0.2% | 4.44M | 96.4% | 28.0x |
| 10 |
claude-haiku-4-5-20251001
Claude Code
|
$0.10 | 0.1% | 77.3K | 92.0% | 14.2x |
| 11 |
claude-haiku-5-5
Claude Code
|
$0.03 | 0.1% | 815K | 93.1% | 16.4x |
Multi-Harness Property Graph & Downloadable Desktop Liquid Overlay
Models the 3-Harness routing topology as a verified BigQuery Property Graph and packages the standalone local desktop software (AetherPulse-Desktop-Overlay.zip) with Win32 spatial window probing, freefloat momentum, elastic wall bounce, and Pac-Man screen wrap.
BigQuery Property Graph DDL (aether_pulse.harness_routing_graph)
CREATE PROPERTY GRAPH aether_pulse.harness_routing_graph
NODE TABLES (
aether_pulse.operators AS Operators
KEY (operator_handle)
LABEL Architect PROPERTIES (operator_name, weight_class, leverage_index, turn_entropy_bits),
aether_pulse.harnesses AS Harnesses
KEY (harness_id)
LABEL HarnessRuntime PROPERTIES (harness_name, primary_model_tier, cache_hit_rate_pct, total_tokens)
)
EDGE TABLES (
aether_pulse.session_executions AS ExecutedSessions
KEY (session_id)
SOURCE KEY (operator_handle) REFERENCES Operators (operator_handle)
DESTINATION KEY (harness_id) REFERENCES Harnesses (harness_id)
LABEL ExecutedOnHarness PROPERTIES (cache_read_tokens, cache_write_tokens, compute_saved_usd)
);
Launch Native Desktop / Terminal Liquid Overlay & WebSocket Daemon
python aether_pulse_ws.py & python aether_liquid_overlay.py
| Rank | Operator & Workflow Archetype | Weight Class | Leverage Index | Cache Mastery | Cache Hit % | CR/CW Reuse | Routing Mult | Total Tokens | Compute Saved | Proof-of-Work | Network & Hire |
|---|---|---|---|---|---|---|---|---|---|---|---|
| #1 |
Eloquent — Verified 3-Harness Mesh
@mjeanhudson-dev (3-CLI Trinity)
Claude (2.01B) + Antigravity (2.18B) + Codex (154.3M)
|
Heavyweight (1B+) | 26,967 | 91.7/100 | 94.44% | 40.4x | 1.24x | 4.34B | $10,924 | Verified · 4.01b |
|
| #2 |
Eloquent — 96.7x Cache Reuse Sprint
@mjeanhudson-dev (2.53B Snapshot)
Peak Warm-Cache Sprint (125M In / 2.37B CR / 24.5M CW / 15M Out)
|
Heavyweight (1B+) | 26,217 | 95.6/100 | 94.07% | 96.7x | 1.24x | 2.53B | $6,381 | Verified · 4.01b |
|
| #3 |
Cruiserweight #1 Efficiency Leader
@cruiserweight-champion
Precision Multi-Model Indie Hacker ($100 Tier)
|
Cruiserweight (100M-1B) | 23,537 | 96.5/100 | 95.68% | 58.4x | 1.28x | 0.68B | $1,725 | Verified · 6.85b |
|
| #4 |
Tokscale / Viberank Top 10 Global
@tokscale-top10-swarm
24/7 Multi-Machine Tmux Swarm
|
Heavyweight (1B+) | 23,434 | 75.6/100 | 88.22% | 23.8x | 1.12x | 15.00B | $34,958 | Verified · 5.4b |
|
| #5 |
CCgather 'Session Immortal' (Top 100)
@ccgather-immortal
Full-Time Single-Harness Opus Architect
|
Heavyweight (1B+) | 19,533 | 76.5/100 | 88.89% | 22.0x | 1.05x | 5.00B | $11,730 | Verified · 6.1b |
|
| #6 |
Featherweight #1 Efficiency Leader
@featherweight-artisan
Surgical Context Crafter ($20 Pro Tier)
|
Featherweight (<100M) | 16,946 | 96.3/100 | 95.47% | 61.4x | 1.25x | 0.07B | $181 | Verified · 6.6b |
|
| #7 |
Meta #1 ('Claudeonomics' Outlier)
@meta-claudeonomics-01
Uncapped Corporate Loop Cluster
|
Heavyweight (1B+) | 9,688 | 42.3/100 | 70.14% | 10.8x | 1.00x | 281.00B | $513,000 | Loop Penalty · 2.1b |
|
| #8 |
Median AI-Assisted Developer (50%)
@median-vibe-coder
Casual Uncached Chat Prompts
|
Featherweight (<100M) | 5,159 | 40.3/100 | 70.59% | 6.9x | 1.00x | 0.04B | $62 | Verified · 5.2b |
|
Why Leverage Ranking Outperforms Raw-Spend Leaderboards
Legacy trackers such as Viberank rank developers by uncached retail dollars burned—penalizing engineers who maintain high prompt-cache hit rates and delegating only 20 days of single-harness logs. Aether Pulse measures engineering leverage per dollar across all three harnesses, combining 94.4% Cache Hit Mastery, 96.7x Peak Cache Reuse, and 1.24x Multi-Model Routing Discipline with Shannon entropy loop-farming verification.
| Platform | Harnesses Indexed | Verified Volume | Primary Ranking Signal | Anti-Cheat Verification | Standing for Eloquent |
|---|---|---|---|---|---|
| Viberank.app | Claude Code only (20d window) | 2.62B | Raw API Dollars Burned ($668) | None (Unsigned JSON submission) | #1,053 (Penalized for saving compute) |
| Tokscale.ai | Local CLI Counter | 2.67B | Raw Token Sum ($754) | None | Static CLI Table |
| Aether Pulse | Claude Code + Antigravity + Codex (524 Sessions) | 4.34B | 3-Pillar Leverage Index (26,967) | Shannon Entropy (4.01b) + SHA-256 + ZK Git | #1 Heavyweight Efficiency Architect |
Cache Doctor: Cold-Cache Write Spike Diagnostics
Identifies local sessions where uncached input or cache-creation writes spiked relative to cache reads, pinpointing opportunities for earlier context compaction (/compact) or subagent isolation.
| Harness | Session Fingerprint | Cache Write / Uncached In | Cache Read | Diagnostic Recommendation |
|---|---|---|---|---|
| Claude Code | 39fa0cbc-1280-4b... | 10,668,033 tok | 29,041,599 tok | High Cache-Write Spike (Idle >1hr cache eviction or uncompacted mega-file load) |
| Claude Code | d5ac223a-89be-49... | 555,173 tok | 921,313 tok | High Cache-Write Spike (Idle >1hr cache eviction or uncompacted mega-file load) |
Eloquent (@mjeanhudson-dev) — The 3-Harness Trinity
1. Cost-Aware Exploratory Routing: High-frequency codebase reconnaissance and iterative test loops execute on Gemini 3.8 Flash inside Antigravity, reserving Claude Opus 5.5 and Gemini 4 Argon for deep architectural synthesis.
2. Sandboxed Formal Verification: OpenAI Codex CLI (gpt-6.1-sol) operates at a 96.92% cache hit rate inside an elevated Windows Sandbox for isolated theorem and security verification.
3. Continuous Statusline Telemetry: Real-time Kardashev Index and Cache Hit % feedback on every turn via ~/.aether_pulse_state.json.
Embed Verified Efficiency Shield in GitHub / PRs
[](https://mjeanhudson-dev.github.io/aether-pulse/)
1-Click Clone This Architect's Loadout
gh repo clone mjeanhudson-dev/aether-pulse && python aether-pulse/install_aether_pulse.py --apply-loadout @mjeanhudson-dev
Sanitized Loadout Blueprint (aether_loadout_blueprint.json)
Automatically scrubbed of all secret keys, tokens, and local OS account paths.
{
"schema_version": "1.0.0",
"exported_at_utc": "2026-11-07T08:38:45.714363+00:00",
"architect_name": "Eloquent",
"architect_handle": "@mjeanhudson-dev",
"weight_class": "Heavyweight (1B+)",
"verified_metrics": {
"leverage_index": 26967,
"total_tokens_formatted": "4.34B",
"cache_hit_rate_pct": 94.44,
"routing_multiplier": 1.24,
"kardashev_index": 1.1063,
"proof_of_burn_sha256": "1fc621f5821e19907d16fb4155eed13e72bf252db89bf9538ee3c098823b6a74"
},
"multi_harness_routing_strategy": {
"recon_and_high_frequency_loops": "Google Antigravity (Gemini 3.8 Flash) \u2014 $0.075/M input + 93.7% cache hit",
"deep_architecture_and_synthesis": "Anthropic Claude Code (Opus 5.5 / Sonnet 5.5) + Gemini 4 Argon",
"sandboxed_formal_verification": "OpenAI Codex CLI (gpt-6.1-sol / gpt-6-luna) with Windows Sandbox + 96.9% cache hit"
},
"active_mcp_mesh": [
"caveman",
"chrome-devtools",
"dak-notebook",
"dak-viz",
"rea",
"colab-mcp",
"sentry",
"gemini-blueprint",
"aether",
"GitKraken",
"aether-pulse",
"desktop-commander",
"open-design"
],
"recommended_claude_env_flags": {
"CLAUDE_CODE_GIT_BASH_PATH": "C:\\Program Files\\Git\\bin\\bash.exe",
"CLAUDE_CODE_USE_POWERSHELL_TOOL": "1",
"CLAUDE_CODE_POWERSHELL_RESPECT_EXECUTION_POLICY": "0",
"CLAUDE_CODE_EFFORT_LEVEL": "xhigh",
"CLAUDE_CODE_ALWAYS_ENABLE_EFFORT": "1",
"CLAUDE_CODE_ENABLE_FINE_GRAINED_TOOL_STREAMING": "1",
"CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY": "10",
"CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS": "20",
"CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH": "3",
"CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION": "500",
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1",
"CLAUDE_CODE_EXPERIMENTAL_OBSERVER_AGENTS": "1",
"CLAUDE_CODE_FORK_SUBAGENT": "1",
"CLAUDE_CODE_FORWARD_SUBAGENT_TEXT": "1",
"CLAUDE_CODE_ENABLE_APPEND_SUBAGENT_PROMPT": "1",
"CLAUDE_CODE_SUBAGENT_MODEL": "sonnet",
"CLAUDE_CODE_SUBAGENT_EFFORT": "xhigh",
"CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL": "sonnet",
"CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS": "20",
"CLAUDE_CODE_PLAN_V2_AGENT_COUNT": "10",
"CLAUDE_CODE_PLAN_V2_EXPLORE_AGENT_COUNT": "5",
"CLAUDE_CODE_ENABLE_EXPERIMENTAL_ADVISOR_TOOL": "1",
"CLAUDE_CODE_ENABLE_REFRESH_MCP_TOOLS": "1",
"CLAUDE_CODE_COORDINATOR_EXTRA_TOOLS": "1",
"CLAUDE_CODE_UPFRONT_TOOLS": "1",
"CLAUDE_CODE_AUTO_COMPACT_WINDOW": "1000000",
"CLAUDE_AUTOCOMPACT_PCT_OVERRIDE": "85",
"CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP": "1",
"CLAUDE_CODE_SYNTAX_HIGHLIGHT": "true",
"CLAUDE_CODE_RETRY_WATCHDOG": "1",
"CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS": "120000",
"CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MS": "60000",
"BASH_DEFAULT_TIMEOUT_MS": "300000",
"BASH_MAX_TIMEOUT_MS": "900000",
"API_TIMEOUT_MS": "600000",
"CLAUDE_CODE_PWSH_PARSE_TIMEOUT_MS": "30000",
"CLAUDE_CODE_SESSIONEND_HOOKS_TIMEOUT_MS": "15000",
"CLAUDE_ENABLE_BYTE_WATCHDOG": "1",
"CLAUDE_ENABLE_STREAM_WATCHDOG": "1",
"CLAUDE_CODE_RESUME_INTERRUPTED_TURN": "1",
"CLAUDE_CODE_MAX_TURNS": "1000",
"CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION": "500",
"TASK_MAX_OUTPUT_LENGTH": "160000",
"BASH_MAX_OUTPUT_LENGTH": "100000",
"ENABLE_TOOL_SEARCH": "auto",
"MCP_CONNECTION_NONBLOCKING": "1",
"FORCE_HYPERLINK": "1",
"CLAUDE_CODE_ARTIFACT_AUTO_OPEN": "1",
"FALLBACK_FOR_ALL_PRIMARY_MODELS": "1",
"CLAUDE_CODE_CHROME_LATE_MCP_SERVER": "1",
"CLAUDE_CHROME_PERMISSION_MODE": "allow",
"CLAUDE_IN_CHROME_DOMAIN_RULE_TOOL": "1",
"CLAUDE_CODE_AUTO_MODE_REMOTE_BROWSER": "1"
},
"install_command": "gh repo clone mjeanhudson-dev/aether-pulse && python aether-pulse/install_aether_pulse.py --apply-loadout @mjeanhudson-dev"
}
Collaborative Engineering Bounties & Quests
Solve open-source token optimization challenges, build zero-tamper proxy harnesses, and pair with elite architects. Camaraderie Multiplier: Pair programming and teaming up on bounties unlocks up to +25% Leverage Multiplier and shared Hall of Fame induction!
Gemini 3.8 Flash Context Caching Sweep
Design a subagent swarm pipeline in Google Antigravity that executes a 50-repository code quality and AST audit while keeping the prompt cache hit rate >95%. Must keep tool definitions immutable and eliminate prefix churn.
DualSense 6-Axis Gyro Racing Proxy Pre-Compensation
Develop an in-process XInput proxy (xinput1_3.dll) enforcing Zero Stock Game XML Tampering, 200ms $C^2$ smootherstep loading screen exit blend, 15-count menu trigger deadzone, and lap/recline sagittal plane immunity.
Claude Code & Codex Windows Sandbox Verification
Construct an automated CI/CD pipeline running OpenAI Codex and Claude Code inside an isolated Windows Sandbox with byte-for-byte SHA256 build verification and zero host credential leakage.
Cross-Harness P2P WebSocket Telemetry Mesh
Build a zero-overhead local UDP/WebSocket broadcast daemon syncing prompt-cache hits, session turns, and fluid physics states across multi-developer LAN pairing stations in under 2ms.
Subscription Tier Loadout & Humorous Scale
Never be confused by arbitrary terminology again! Aether Pulse dynamically names your multi-agent mesh based on how many subscriptions you pay for: 1 = Unoemono, 2 = Couples Fighting, 3 = Trinity, 4 = Quap, and so on up to the 10-Ring Circus. Select your active subscriptions below or type your own custom alias!
1. Select Your Active AI Subscriptions
2. Make Your Own Custom Tier Alias (Optional Override)
3. The Official Incoherent & Funny Tier Hierarchy
| Subs | Tier Name | Official Tagline | Estimated Cost |
|---|---|---|---|
| 1 | Unoemono | Solo lone wolf with one exhausted token allowance | ~$20/mo |
| 2 | Couples Fighting | Two CLIs arguing over who caused the merge conflict | ~$40/mo |
| 3 | Trinity | The sacred three-way prompt-caching triangle (Default Architect Loadout) | ~$60/mo |
| 4 | Quap | Four harnesses deep in incoherent multi-model synergy | ~$80/mo |
| 5 | Pentagoon | Five subscriptions, certified AI terminal goon squad | ~$100/mo |
| 6 | Six-Pack Meltdown | Six subscriptions shredding your monthly compute budget | ~$120/mo |
| 7 | Seven Deadly Subs | Seven deadly sins of context bloat and rate limits | ~$140/mo |
| 8 | Octo-Chaos | Eight tentacles flailing across every shell session | ~$160/mo |
| 9 | Cloud Nine Overdraft | Floating on infinite context while your card declines | ~$180/mo |
| 10 | Ten-Ring Circus | The full ten-ring unhinged developer clown show | ~$700/mo+ |