Benchmarks
Resurface vs a general LLM.
Resurface is built on TypeSafe Jev, a decision model made for exactly this kind of question, and adds everything around it: file parsing, screening questions, quoted evidence, a fallback model and an eval. Here it is measured against general LLMs on the same 20 fictional resumes and 5 screening questions for a Senior Full Stack Engineer, with Resurface's time taken from real production requests.
Resurface API
end to end, powered by Jev
628 ms
Claude Haiku 4.5
general LLM, model only
1.7 s
Claude Sonnet 5.5
general LLM, model only
2.5 s
Resurface API
our production request
99%
Claude Haiku 4.5
general LLM
93%
- Resurface API time is the full request in production, including file parsing, sign-in, billing and storage: the median of all 94 production scores (1 Oct 2026–2 Oct 2026). Requests for a new job description add a few seconds once, to draft the questions.
- LLM times (model only) and the Haiku agreement come from the open-source benchmark, 2 runs each, 30 Sep 2026. Sonnet is the reference for agreement, so it isn't plotted there.
- Resurface agreement is our exact production request on the same sample, 2 Oct 2026.
What each approach gives you
| Resurface API · powered by TypeSafe Jev | A general LLM | |
|---|---|---|
| Time per resume | 628 ms end to end | 1.7 s–2.5 s model only |
| Same answer on a re-run | 100% | 98% (Haiku) · 100% (Sonnet) |
| Resume files | PDF, DOCX, HTML, TXT, two-column layouts in reading order | You send text or a file |
| Screening questions | Drafted from the job description once, then reused | Per prompt, every call |
| Evidence for each answer | The resume line, copied by design | Generated text you must check |
| Years of experience | Computed from dated roles, overlaps merged | Estimated by the model |
| If the model is down | Automatic fallback to Cloudflare Clef-flash | You build it |
| Before a model serves scores | Must pass our resume eval | You build it |
| Inside Claude and MCP clients | Plugin and MCP server | You build it |
| Price per resume | $0.06–$0.10, all included | Token costs, plus the pipeline you build and host |
Run it yourself. The resumes, the questions and the benchmark script are open source. Resurface's speed and consistency come from TypeSafe Jev; we build the screening pipeline around it. Try it with 50 free scores.