In Zika Zag’s author-run test of 53 tasks with known projects, Claude Opus 5 made more correct assignments than Jev 1.13: 33 versus 30. Jev made fewer wrong assignments (9 versus 12) and left more tasks unassigned (14 versus 8). The distinction matters in Open Walnut: a wrong project can hide a task from the list a user checks, while a blank leaves it in the Inbox. Zag also reported much lower median latency for Jev—209 ms versus 4,689 ms for Opus—but this small, single-author case study is not a general model ranking. Read Zag’s original account on DEV Community.
What the 53-task comparison found
Zag evaluated project assignment for 53 personally selected tasks whose correct projects were already known. The following figures are reported by the author, not independently verified. “Right when answered” excludes blanks; a blank counted as neither right nor wrong.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
As an Amazon Associate I earn from qualifying purchases.
| Model and route | Right | Wrong | Blank | Right when answered | Median latency | Cost per call |
|---|---|---|---|---|---|---|
| Jev 1.13 through OpenRouter | 30 | 9 | 14 | 77% | 209 ms | $0.00019 median, based on 53 calls |
| Claude Opus 5 through Bedrock | 33 | 12 | 8 | 73% | 4,689 ms | About $0.025 estimated; see calculation below |
| Claude Sonnet 4.6 through Bedrock | 29 | 15 | 9 | 66% | 1,758 ms | Not measured; author’s article |
| Claude Haiku 4.5 through Bedrock | 19 | 8 | 26 | 70% | 1,013 ms | Not measured; author’s article |
These are results for project assignment, not a combined score across project, priority, tier, or session moves. The row totals and conditional accuracy show why the outcome depends on what matters most: Opus produced the most correct assignments, while Jev made fewer mistakes and abstained more often. A wrong filing and an unanswered filing have different consequences in this app.
Why a blank can be safer than a wrong project
In Open Walnut, a task assigned to the wrong project may disappear from the list the user expects to check. A blank instead remains in the Inbox. That makes abstention a deliberate trade-off: the application gives up automatic sorting when it is uncertain to reduce the risk of misplaced work.
#1 Best Overall
Jev’s confidence floor controls that choice. In the described setup, the project floor was 0.4; tier and priority used 0.5; session moves used 0.6. A field below its floor was left blank. Zag reports that lowering the project floor from 0.5 to 0.4 added nine right assignments and one wrong assignment in this set. In a production-path comparison, the earlier wording and 0.4 floor yielded 22 right, 8 wrong, and 23 blank; the revised wording and same floor yielded 30 right, 9 wrong, and 14 blank. These are author-reported outcomes from this task set, not a controlled finding that the threshold or wording will improve other users’ results.
How the application routes and scores a decision
What Jev receives and returns
The described Open Walnut workflow uses Jev for two filing tasks: suggesting a project, priority, and tier while someone types in a draft composer, and deciding whether a quick-start session should move into a project. Rather than generating free-form prose, Jev receives a text state containing the note and project digest, along with typed choice questions. Its response supplies a selected option, option probabilities, and confidence.
How project choices are represented
The digest lists the 20 projects with the most open tasks, while the choice question offers every project. Zag says that including each project’s short summary—up to 160 characters—as evidence for its option helped represent projects outside the top-20 digest. This is a detail of the implementation, not proof that the same prompt design will work equally well elsewhere.
Scoring rules and ambiguous tasks
A blank counted as neither correct nor incorrect. Zag also treated three notes that seemed equally compatible with two projects as wrong for every engine. That convention makes the comparison easier to interpret, but it is a judgment call within a small labeled set rather than an objective resolution of those ambiguous tasks.
Latency and cost are not equally measured
The author reports median latency of 209 ms for Jev, 4,689 ms for Opus, 1,758 ms for Sonnet, and 1,013 ms for Haiku under the described production paths. These are measurements from Zag’s setup; the article does not establish that other regions, accounts, prompts, or service conditions will produce the same timings.
The cost figures are less directly comparable. Jev’s $0.00019 is the median reported `usage.cost` across 53 calls. Opus’s approximately $0.025 is an estimate based on five calls, roughly 4,290 input tokens and 130 output tokens, token prices cited as $5 and $25 per million respectively, and no prompt caching. Sonnet and Haiku costs were not measured. The estimates should not be read as an audited, like-for-like cost benchmark.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Reliability is different from accuracy
In a separate run of 92 unlabelled to-dos, Zag reports that all calls returned, with a median response time of 162 ms and a 95th percentile of 489 ms. Because those tasks had no known correct answers, the run speaks to successful call returns and response timing in that instance—not project-classification accuracy.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteThe implementation also distinguishes an intentional abstention from a failure. Low confidence or a “nothing fits” response leaves the field blank. Network errors, malformed responses, or missing confidence instead trigger a fallback to the prior fast-model path. A 2.5-second limit applies to the quick-add call. This behavior is specific to the described Open Walnut integration.
Best Value
What the comparison can—and cannot—establish
The evidence is a single author’s test of 53 known-project tasks in one application path. Jev’s project assignments, confidence cutoffs, prompt wording, task mix, and routing differ from a generic head-to-head model test; the results do not establish a universal winner between Jev and Claude. The table is most useful as a case study in how accuracy, mistakes, abstentions, latency, and cost can pull in different directions when automating task filing.
For the implementation details and author’s full account, see Zika Zag’s September 24, 2026 article on DEV Community.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
