Compare
Cost per request, same task, same output. Provider prices are official list rates; the Intelysis bar is our measured engine result.
Request sizecoding agent, repo context · 41,000 in / 2,000 out
Intelysis vs direct
6.67×
cheaper, same output
Cost / request
$0.0011
vs $0.0073 direct
vs cheapest frontier
9.6×
vs GPT-5.6-luna
vs premium frontier
408×
vs Claude Fable 5.1
Cost per request (USD, lower is better)
41,000 input · 2,000 output tokens
Intelysis (our engine)
$0.0011baseline
DeepSeek V4.1-Flash
$0.00736.7×
GPT-5.6-luna
$0.0119.6×
Claude Haiku 4.5
$0.04540.8×
Claude Sonnet 5
$0.09081.6×
GPT-5.4
$0.133120.2×
Claude Opus 5
$0.225204.1×
Claude Fable 5.1
$0.450408.2×
Prompt-compression toolsreported
$0.00494.4×
Intelysis optimizedverified published pricereported, not measured by us
Data & sources
| Model / engine | Input $/M | Output $/M | Cost / request | Basis |
|---|---|---|---|---|
| Intelysis (our engine) | 0.150* | 0.60 | $0.0011 | measured · 85.0% lower bill |
| DeepSeek V4.1-Flash | 0.15 | 0.60 | $0.0073 | official list price ↗ |
| GPT-5.6-luna | 0.20 | 1.20 | $0.0106 | official list price ↗ |
| Claude Haiku 4.5 | 1.00 | 2.00 | $0.0450 | official list price ↗ |
| Claude Sonnet 5 | 2.00 | 4.00 | $0.0900 | official list price ↗ |
| GPT-5.4 | 2.50 | 15.00 | $0.1325 | official list price ↗ |
| Claude Opus 5 | 5.00 | 10.00 | $0.2250 | official list price ↗ |
| Claude Fable 5.1 | 10.00 | 20.00 | $0.4500 | official list price ↗ |
| Prompt-compression tools | 0.150* | 0.60 | $0.0049 | reported range · not measured by us |
* Intelysis bills the underlying DeepSeek input rate on the prompt it actually sends. DeepSeek off-peak list price shown; peak is 2×. The engine row applies the measured bill from the cold coding suite of 2026-11-12: $0.078246 → $0.011702, 85.0% lower, with every task passing in both arms. Token reduction was 86.3%; it is not the bill number because output tokens and cache hits do not shrink with the prompt. Reproduce it on your own traffic in Benchmarks.