A Stanford and Tsinghua paper ran a controlled experiment earlier this year. Same model. Same task. Different harness architecture.
The result: a 6x performance gap driven entirely by the system built around the model. Not the model itself.
This is not a prompt engineering insight. It is a systems architecture insight, and it changes where developers should invest their time when building agentic systems.
The 6x Gap
Meta-Harness tested Claude Opus 4.6 across two harness configurations on TerminalBench-2. The only variable was the scaffold: the code that manages tool calls, context windows, error recovery, and state persistence.
One version scored at baseline. The other, with structured tool orchestration and context management, scored 18.4 points higher. Same inference cost. Same model. Different architecture.
This pattern replicates across multiple independent studies:
LangChain DeepAgents (2026): Same GPT-5.2-Codex model. Harness-only changes moved it from Top 30 to Top 5. That is a 13.7-point gain.
Can Bölük (Hashline, 2026): Same model, same task. Changed the edit tool format. Performance went from 6.7% to 68.3%. That is a 10x improvement with 61% fewer tokens.
Vercel's d0 agent: A production agent had 16 tools. Removing 14 of them (leaving only bash) took success rate from 80% to 100%. The bottleneck was not capability. It was decision surface.
Why This Matters Practically
The cheapest Haiku call with an optimised harness (37.6% on TerminalBench-2) outperformed the most expensive Opus call with a default harness (58.0%). That is at 1/50th the inference cost.
Most teams are optimising at the wrong layer. They swap models, tune prompts, add retrieval. The structural leverage is in how the system manages tool calls, handles state, and recovers from failure.
What Changes
The practical takeaway for anyone building with AI agents:
Audit your tool surface. Every tool your agent can call is a decision it must make. Vercel found 16→1 tool reduction improved everything. Fewer tools, better decisions.
Measure harness, not just model. Track task completion rate per harness configuration, not just per model. The harness is the variable that moved 6x.
Cost is architecture-dependent, not model-dependent. Haiku with a good harness beat Opus with a bad harness. Test harness variations before upgrading to a more expensive model.
The full analysis (12 verified claims, evidence tables, production case studies, and falsification criteria) is on Substack:
Harness Engineering: Same Model, Different Product →
It covers the Claude Code 1,421-line state machine, the Codex CLI vs Claude Code architecture comparison (77.3% vs 65.4%, 4.2x token efficiency difference), and why this is a Law IV (Instruments Over Theory) and Law I (Bottleneck Migration) structural play.
Follow for weekly analysis on AI infrastructure, agent architecture, and the systems that actually determine model performance.