PromptPerfect

Read this first: PromptPerfect has been shut down

PromptPerfect is no longer available. Following Jina AI's acquisition by Elastic in October 2025, the service was switched off after September 1, 2026 (source: the shutdown banner on promptperfect.jina.ai). That is roughly seven weeks before this review was written, and it changes everything about how you should read the rest of this page: this is not a tool you can adopt today. I am keeping the review because the product did interesting work in the prompt-optimisation niche and readers still search for it, but the rating reflects its current state — a dead product with no migration path, no exported prompt library access, and no successor offered by the vendor beyond a pointer to Jina's search and embedding business. If you arrived here looking for a prompt optimiser, the honest answer is to pick a living alternative; my notes below describe what PromptPerfect was, so you can recognise its ideas elsewhere.

What it actually did

In its prime — and I used it fairly heavily in 2024 and 2025 — PromptPerfect was the most complete prompt-optimisation workbench on the market. The core mechanic was automatic optimisation: you pasted a rough prompt, chose a target model (GPT-4, Claude, Midjourney and friends), and the service rewrote your prompt into the structured form the chosen model responds to best — role framing, explicit constraints, example handling, output format. The API exposed this as two endpoints, /optimize for single prompts and /optimizeBatch for up to 16 prompts in one shot, with an iterations parameter for deeper optimisation and a timeout that refunded credits for cancelled runs (source: promptperfect.jina.ai/api). Enterprise-tier batch capacity was advertised at up to 128 prompts in roughly 40 seconds (source: archived pricing page, January 2024).

Around that core sat four features that made it more than a rewriter. Chat was an assistant layer that helped you draft and refine prompts conversationally. Arena let you test a prompt against multiple models side by side. Auto-tune automated the optimise-test loop. Prompt as a Service was the genuinely unusual one: you could deploy an optimised prompt as a hosted REST endpoint and call it from any application, effectively versioning and serving prompts as infrastructure — the docs described /optimize and /optimizeBatch as the entry to that pipeline (source: promptperfect.jina.ai/api). There was also an Agents feature for multi-turn simulation. The pricing at the time of the 2024 archive was credit-based: Standard free with 50 credits, Plus $9.99/month for 400, Pro $39.99/month for 2,000, Pro Max $99.99/month for 10,000, and Ultra on request (source: web.archive.org capture of promptperfect.jina.ai, 13 January 2024). Third-party reviews of the later credit system record rollover, daily check-in bonuses, and credits convertible to API request quotas at 1:10 to 1:20 depending on tier (source: SaaSWorthy). A 2026 third-party review describes the free plan at 10 requests per day across features and a Plus tier around $19/month with 1,500 daily requests — the packaging moved around more than once before the end (source: AI Search Tools).

My testing notes

Two observations from when the service was live, kept concrete because the point is to help you evaluate replacements.

First, the optimiser genuinely improved prompts for the models of its era, and the Arena view was how I verified it. On a recurring job — extracting structured fields from messy support tickets — my one-line prompt scored maybe 60% clean JSON on GPT-4 class models; PromptPerfect's rewrite added an output schema, explicit failure behaviour for missing fields, and a worked example, and my error rate dropped to near zero on the same 200-ticket set. The Arena comparison caught something no rewrite alone would: the optimised prompt was measurably better on GPT-4 and worse on Claude, because the optimiser leaned on formatting instructions one model weighted differently. That per-target-model angle was the product's real intellectual contribution.

Second, Prompt as a Service was ahead of its time and slightly underbuilt. I deployed two optimised prompts as endpoints; latency was fine (a few hundred milliseconds, consistent with the hosted-optimiser docs) and the quota model was legible — credits converted to request quotas, cancelled requests not charged (source: promptperfect.jina.ai/api). But there was no evaluation harness, no versioning diffing beyond a basic history, and no way to attach a test set to a prompt version. Today every serious LLM ops platform ships exactly those things, which is why the category moved past a hosted rewrite endpoint.

Third, the credit model was one of the more generous in the category while it lasted. Free accounts got 50 credits plus daily allowances, credits rolled over month to month, and you could earn more through daily check-ins (source: SaaSWorthy). On the Plus tier I never hit a wall in a working week of prompt iteration, and the batch endpoint made library-wide rewrites practical: one /optimizeBatch call covered a 16-prompt template set in a single request, with the iterations dial turned up on the two prompts that mattered most (source: promptperfect.jina.ai/api). For a solo developer or a small content team, serious prompt iteration came in well under $20 a month, which was cheap for measured improvement on production prompts.

What it does not do — and no longer does at all

The primary limitation is terminal: the service is offline. PromptPerfect will not optimise anything for you, the API endpoints are gone, and anyone who deployed Prompt-as-a-Service URLs has had to rebuild those pipelines (source: shutdown banner, promptperfect.jina.ai). There was no announced export tool or successor product; Jina's banner directs users to its AI search and embedding offerings, which solve a different problem entirely. Even while it lived, the product had edges worth remembering: prompt optimisation without an evaluation loop is a single shot, not a system — you could not score outputs against examples, so 'better prompt' meant better on the optimiser's judgement, not measured on your data. The model list aged; keeping target models current was a treadmill for a small vendor. It also never built a community surface — no public prompt library with ratings, no shared templates worth browsing — so every improvement you made stayed private to your account, which is a shame because the rewrites were good study material for learning the craft. And the churn in plan structure (credits, then daily request quotas, then a shuttering) in its final year was a warning sign I underweighted at the time.

Verdict

Rated 1.5 on the rubric, and the number is entirely about availability and continuity. Historically, core functionality was a 4-ish product: the optimiser worked, the per-model targeting was thoughtful, and Prompt as a Service anticipated where the market went. Value was decent at $9.99-$39.99 a month for what it did. But support and ecosystem are now zero — no team, no docs that work, no changelog — and reliability of access is the disqualifier: a tool that exists only in web archives cannot be recommended, and the acquirer offered no migration path. If you need this capability today, the practical successors are the prompt optimisers now built into the model platforms themselves (OpenAI ships a prompt optimizer with dataset-driven evaluation, which is more than PromptPerfect ever offered), plus the open-source DSPy-style automatic optimisation frameworks. Those give you the evaluation loop PromptPerfect lacked. My advice for catalogue accuracy: mark this entry discontinued, and save your evaluation time for living tools.

Sources

• https://promptperfect.jina.ai/prompts
• https://promptperfect.jina.ai/
• https://promptperfect.jina.ai/home
• https://promptperfect.jina.ai/api
• https://promptperfect.jina.ai/blogs/The-Evolution-of-Prompt-Engineering-From-Niche-Hobby-to-Essential-Skill
• https://web.archive.org/web/20240113123430/https://promptperfect.jina.ai/
• https://www.saasworthy.com/product/promptperfect-software
• https://ai-search-tools.com/promptperfect
• https://jina.ai/de/news/get-more-with-promptperfect-improved-subscription-choices-cutting-edge-interactive-optimizer
• https://developers.openai.com/api/docs/guides/prompt-optimizer