Published
Updated
Polylane moved five on-call decisions to Jev and prints a 39.1% lower bill per 1,000 calls
This topic was created 31 days ago, and the information it contains may have evolved or changed since then.
Vi Tran posted Polylane's write-up on September 28. The page, dated September 25, replaces five fixed-answer steps with jev-1.13.0. The rollup is p90 latency 4,752 ms to 508 ms, and $0.76199 to $0.46369 per 1,000 calls. The page does not print accuracy or a call count. We did not rerun the traffic.
Vi Tran posted on September 28 that Polylane had swapped language-model calls for Jev on the small decisions inside its on-call agent. The write-up is by Tran and Boris Tane. The page’s own date is September 25. The samples name jev-1.13.0. The questions are closed: should the agent answer, how two incidents connect, why a pull request was closed, how pressing a fix is, and which cloud resource matters.
The page prints five before-and-after pairs. Routing, previously DeepSeek V4.1 Flash, moves from a p90 of 1,466 ms and $0.238 per 1,000 calls to 472 ms and $0.123. Incident connection, previously GPT-OSS 120B, moves from 2,859 ms and $0.388 to 373 ms and $0.285. The reason a pull request was closed, also GPT-OSS 120B, moves from 2,379 ms and $0.123 to 416 ms and $0.102. Autofix severity, DeepSeek again, moves from 1,456 ms and $0.197 to 313 ms and $0.081. Ranking a cloud resource, GPT-OSS 20B, moves from 5,445 ms and $0.817 to 511 ms and $0.542. The page says ranking is the highest-volume decision and drives most of the savings. Routing is the only row it dates, as about a week of traffic.
The summary box blends those rows: p90 latency 4,752 ms to 508 ms, which it calls 89.3% faster, and $0.76199 to $0.46369 per 1,000 calls, which it calls 39.1% cheaper. A later chart then prints one point per model, and those points are not the task rows. DeepSeek V4.1 Flash is 1,372 ms and $0.51 there, while the two DeepSeek task rows are 1,466 ms and 1,456 ms. GPT-OSS 120B is 2,255 ms and $0.74. Jev is 508 ms and $0.46. The page does not print the weights, the call counts, or a date range for the blend.
Nothing on the page says whether the new answers matched the old ones, or matched a person. The sample JSON shows the question shape. It is not a scored set. We did not see the production logs.
This site's reading
Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.
Verify
Post by @viviane_trn, September 28, 2026, 22:18 UTC. The page it links is by Vi Tran and Boris Tane. Schema.org dates the article September 25. The company line on the page is Coreplane Labs.
Five production comparisons, model id jev-1.13.0 in the sample JSON. Routing against DeepSeek V4.1 Flash: p90 1,466 ms to 472 ms, $0.238 to $0.123 per 1,000 calls. Incident connection against GPT-OSS 120B: 2,859 ms to 373 ms, $0.388 to $0.285.
Pull-request close reason against GPT-OSS 120B: 2,379 ms to 416 ms, $0.123 to $0.102. Autofix severity against DeepSeek V4.1 Flash: 1,456 ms to 313 ms, $0.197 to $0.081. Resource ranking against GPT-OSS 20B: 5,445 ms to 511 ms, $0.817 to $0.542.
The summary row is p90 4,752 ms to 508 ms, 89.3%, and $0.76199 to $0.46369 per 1,000 calls, 39.1%. A later chart prints different per-model rollups: GPT-OSS 20B 5,445 ms and $0.82, DeepSeek V4.1 Flash 1,372 ms and $0.51, GPT-OSS 120B 2,255 ms and $0.74, Jev 508 ms and $0.46.
Routing is the only comparison the page dates, "about a week." No call counts, no accuracy, no disagreement rate. We did not see the logs.
Compare
Deel's production chart prints quality deltas and a cost multiple. This page prints latency and dollars for five closed questions and does not say whether the labels stayed put. LangChain's judge note is 500 matching pass/fail labels.
Polylane's 39.1% is a blended bill per 1,000 calls, not a quality result.
Terms
- 39.1%
- The page's reduction from $0.76199 to $0.46369 per 1,000 calls, across the five decisions it rolled up. The same box prints p90 latency from 4,752 ms to 508 ms, 89.3%.
- $0.542
- Jev's cost per 1,000 calls on resource ranking, against GPT-OSS 20B at $0.817. p90 on that row is 511 ms against 5,445 ms. The page calls this the highest-volume decision.
- 1,372 ms
- DeepSeek V4.1 Flash on the page's later per-model chart. The two DeepSeek task rows above it print 1,466 ms and 1,456 ms. The page does not say how the 1,372 ms rollup was weighted.