Run Claude Code, pi, or delegated coding subagents on a 3-tier open-weight ladder
(qwen3.6-35b-fast → deepseek-v4.1-flash → kimi-k3.1-pro)
paired with static code-graph slicing and deterministic architecture checks. Compare live NVML GPU Joules
against Claude 5.5 and 3T Frontier MoE baselines.
Your API key below works immediately in Claude Code, pi, any Anthropic- or OpenAI-compatible coding agent,
or the optional ecocoder repo helper. Every turn logs NeuralWatt's metered Wh and enforces your $25.00 cap.
| Counterfactual Model (2026 Priors) | Assumed Active / Total | Method A: GPU Wh [Opt .. Cons] | Method B: EcoLogits 0.11 Wh | Multiple vs. EcoCoder | Energy Saved |
|---|---|---|---|---|---|
| Loading counterfactual model physics... | |||||
ANTHROPIC_BASE_URL="https://api.leanlm.ai" \ ANTHROPIC_API_KEY="<LOADING_KEY>" \ claude
OPENAI_BASE_URL="https://api.leanlm.ai/v1" OPENAI_API_KEY="<LOADING_KEY>" # Model ID: claude-sonnet-4-5 (or ecocoder)
./ecocoder for Repo Code-Graph Slicing, pi Setup & Subagentscurl -fsSLO https://api.leanlm.ai/ecocoder && chmod +x ecocoder ./ecocoder login --key "<LOADING_KEY>" ./ecocoder stats
cd /path/to/your/repo /path/to/ecocoder claude
cd /path/to/your/repo /path/to/ecocoder pi # Or persist provider: /path/to/ecocoder setup-pi
/path/to/ecocoder install-subagent /path/to/ecocoder delegate \ "Fix edge cases in parser.py" \ --max-usd 1.00
| Key ID | Label | Status | Created | Action |
|---|---|---|---|---|
| No CLI keys loaded. | ||||
Passing a test suite once is cheap if the model dumps untyped dict bags across modules, clones dispatch branches,
or swallows failures behind broad except Exception: blocks. On the next feature request, every model has to pull
3×–5× more files into context to trace hidden side effects, retries broken callers, and escalates to frontier weights.
EcoCoder pairs its 3-tier ladder with codegraph-slicer v0.3.10 (AST call/import blast-radius slicing) and sub-40ms
static checks across four brownfield defect classes: Data-Model (DM001–004), State-Machine (SM001–004),
Evolution-Regression (ER001–004), and Error-Contract (EC001–004).
train5 Brownfield Checkpoint Evolution, 522 Tests)| Scenario | Served Stack | Multi-CP Pass Rate | EC Violations | Retail Token USD | GPU Energy | Joules / Pass | Provenance |
|---|---|---|---|---|---|---|---|
| leanlm-eco (EcoCoder v2) | qwen3.6-35b-fast → deepseek-v4.1-flash → kimi-k3.1-pro | 495 / 522 (94.83%) | 0 | $3.22 | 486.3 kJ (135.1 Wh) | 982 J | NVML metered (5/5) |
| commercial-base | claude-opus-5-5 (unrouted) | 510 / 522 (97.70%) | 0 | $7.95 | 1,328.7 kJ (369.1 Wh) | 2,605 J | Modeled token physics |
| commercial-eco | Claude Haiku / Sonnet / Opus 5.5 | 487 / 522 (93.30%) | 0 | $13.35 | 2,132.8 kJ (592.4 Wh) | 4,379 J | Modeled token physics |
| neuralwatt-base | glm-5.3 (single open-weight) | 464 / 522 (88.89%) | 0 | $6.93 | 2,755.2 kJ (765.3 Wh) | 5,938 J | NVML metered (5/5) |
batch=64 while multiplying per-GPU energy across all weight-holding GPUs; how should effective cluster batch size scale above 16 GPUs?O(L · d²) while KV-cache bytes under GQA/MLA scale sublinearly (~P_act^(2/3) or ~P_act^(1/2)); what exponent best matches long-context HBM traffic?30B / 200B), Sonnet 5.5 (110B / 1.2T), and Opus 5.5 (250B / 2.4T) well-bracketed?