SSimpleTI
Evidence laboratory

Numbers need a cohort before they need a rank.

A real model creates a file and edits it in the same session. The result is checked outside the session: pages in headless Chrome, scripts by running them. 138 sessions, medians.

Claude Sonnet 5.5 · Claude Code · Page from a JSON file, a generator covers itMeasured
No MCP33.4 s
Simplicio 3.10.817.5 s

The gain comes from the generator. On a task no generator covers, Simplicio can cost more than no MCP at all, and the chart shows it.

ModelTaskSetupnTimeModel requestsOutput tokensCostDelivered
Claude Sonnet 5.5 · Claude CodePage from a JSON file, a generator covers itSimplicio 3.10.8517.5 s81,270$0.06725/5
No MCP533.4 s55,147$0.09275/5
New page, other data, other language, not used to build the generatorSimplicio 3.10.8516.3 s81,084$0.05145/5
No MCP522.5 s52,819$0.06105/5
Python script, no generator covers itSimplicio 3.10.8519.6 s71,582$0.05485/5
No MCP511.4 s5644$0.02835/5
Grok 4.7 · Grok BuildPage from a JSON file, a generator covers itSimplicio 3.10.8331.0 s81,289$0.16903/3
No MCP3359.7 s1928,247$0.85903/3
New page, other data, other language, not used to build the generatorSimplicio 3.10.8336.0 s81,598$0.20593/3
No MCP3387.5 s2227,598$0.85183/3
Python script, no generator covers itSimplicio 3.10.8382.0 s115,510$0.33153/3
No MCP363.1 s94,009$0.24523/3
Claude Haiku 5.5 · OpenCodePage from a JSON file, a generator covers itSimplicio 3.10.8513.7 s6918$0.00465/5
No MCP529.3 s56,917$0.00835/5
New page, other data, other language, not used to build the generatorSimplicio 3.10.8513.8 s6794$0.00235/5
No MCP528.7 s56,310$0.00575/5
Python script, no generator covers itSimplicio 3.10.8519.2 s71,583$0.00335/5
No MCP516.6 s71,358$0.00305/5
DeepSeek V4.1 Flash · OpenCodePage from a JSON file, a generator covers itSimplicio 3.10.81014.1 s6602$0.002110/10
No MCP10176.3 s32.517,929$0.020310/10
Simplicio 3.9.210166.2 s24.519,944$0.021510/10
New page, other data, other language, not used to build the generatorSimplicio 3.10.8614.1 s6524$0.00106/6
Simplicio 3.9.26110.9 s17.516,091$0.01536/6
Python script, no generator covers itSimplicio 3.10.8630.7 s10.51,796$0.00316/6
No MCP622.7 s81,268$0.00246/6
Simplicio 3.9.2634.1 s11.52,485$0.00516/6

Each model runs in one fixed setup, the same on both sides: Claude Code, Grok Build, or OpenCode through OpenRouter. Cost is the list price of the tokens used. These are small tasks: other tasks will give other numbers.

MetricRequirement for claimCurrent state
Ranking against other tools10166.2 s24.519,944$0.021510/10
No MCP10176.3 s32.517,929$0.020310/10
New page, other data, other language, not used to build the generatorSimplicio 3.10.8614.1 s6524$0.00106/6
Simplicio 3.9.26110.9 s17.516,091$0.01536/6
Python script, no generator covers itSimplicio 3.10.8630.7 s10.51,796$0.00316/6
Simplicio 3.9.2634.1 s11.52,485$0.00516/6
No MCP622.7 s81,268$0.00246/6

OpenCode 2.1.4 and DeepSeek V4.1 Flash on one fixed OpenRouter endpoint, the same for every run. Cost is computed at $0.15 input, $0.60 output and $0.003 cache read per million tokens. This is one model on three small tasks: other tasks will give other numbers.

MetricRequirement for claimCurrent state
Ranking against other toolsSame task, model, hardware, n and quality thresholdN/A
Token economy1 - tokens_with/tokens_baseline, with cohort and provider priceMeasured
Route smokeVersioned static build and local HTTP checkMeasured locally