Context Engineering Kit
Cost Calculator · v1.0 · Apr 2026
Your Estimated Monthly Savings — with persistent context + prompt caching
$0/mo
That's $0/year you keep by treating context as infrastructure, not chat history.
0% vs. baseline
Scenario A · Baseline
No context file
Re-explain project every session. Extra clarification turns eat tokens on both sides.
$—
/ month
Input tokens
Output tokens
Overhead
+50% tokens
Scenario B · Midway
.md context, no caching
Stable prefix loaded every session, but no API-level cache discount applied.
$—
/ month
Input tokens
Output tokens
vs. Scenario A
Scenario D · Maximum
All optimizations applied
Toggle the optimization levers below to see how low you can go.
$—
/ month
vs. Scenario A
vs. Scenario C
Active levers
0 / 4
Annual savings
$0
Scenario A → C, at your current settings
Per-agent savings
$0
Monthly savings divided across your agents
Sessions this month
0
Total across all agents in 30 days
Optimization Levers
How to save even more
Auto-recommended based on your current setup. Toggle each lever to see its impact on Scenario D.
Additional savings from active levers $0/mo
01
Model downgrade
Use a cheaper model tier for the same work
Currently
Recommended
Saves /mo
02
Batch API (async)
50% discount for non-real-time workloads
Async-tolerant workload 50%
Saves /mo
03
Context compression
Trim redundancy in your .md to cut cached token volume
Currently
Target
Saves /mo
04
Output hard cap
Limit max_tokens per response to prevent runaway outputs
Current avg
Cap at
Saves /mo
Pricing reference · per million tokens · April 2026
Model Input Output Cache Read Cache Write
Methodology
Scenario A assumes the user re-types partial context each session and triggers clarification turns, modeled as a 50% token overhead on input and 20% on output. Scenario B loads a persistent .md prefix but without API-level caching. Scenario C enables Anthropic-style prompt caching (cache writes at 1.25× input, cache reads at 0.10× input), with one cache write per agent per day and the rest amortized across reads. Actual savings vary by cache hit rate, session burstiness, and model-specific caching policies. Pricing compiled from official Anthropic, OpenAI, and Google pricing pages as of April 2026.

Part of the Context Engineering Kit. MIT licensed.