Claude 5 for Job Search: Opus, Sonnet, Fable

Anthropic did in 2026 what needed doing for a while: made the Claude 5 family actually legible. There's Opus, Sonnet, and Fable. And each model genuinely does something different, not just faster or slower, but *different*. For job search that distinction matters: pick the wrong model for the wrong task and you'll either waste time or end up with output you have to redo from scratch.

What separates the three models

Worth understanding what you're actually choosing between. Not benchmarks, just the practical point.

• Claude 5 Opus - the heaviest model. Takes longer, but genuinely analyzes structure. Spots contradictions, logic gaps, weak phrasing. Costs more per token, runs slower. Use it when the task is singular and hard.
• Claude 5 Sonnet - the middle option in the best sense. Fast, sharp, holds context well across messages. Handles iterative work well: 'rewrite this', 'now shorter', 'once more'. The right pick for daily back-and-forth coaching.
• Claude 5 Fable - a model trained specifically on narrative and tone. It doesn't just write text, it sounds like something. Slightly slower than Sonnet, but if you need a letter that reads like a human wrote it on purpose, the choice is obvious.

Opus: when the CV needs surgery, not a face cream

Most people ask AI to 'improve the CV'. And they get a CV where the words got prettier but the actual problem is still there. Opus is for tasks where something needs to be understood, not just rewritten.

Before you sit down with Opus, it helps to already have a baseline score from the AI CV Analyzer. Opus isn't hunting for obvious formatting errors, it's busy with something else.

Exact prompts for Opus

1. 1Career narrative logic. "I'm moving from [X] role to [Y] role. Here's my CV. Find three places where this transition looks illogical or unclear to a recruiter. Explain why, and give a specific rewrite for each."
2. 2Gap audit. "I have a gap in my CV from February 2022 to October 2022. Here's what I was doing: [description]. What's the best way to address this in both the CV and cover letter, so it reads honest but not weak?"
3. 3Position fit. "Here's the job description: [JD]. Here's my CV. List 5 specific mismatches between what they're looking for and how I present myself. For each one, give a recommendation."
4. 4Metrics audit. "Review every bullet point in the experience section. Mark which have specific numbers and which don't. For each without numbers, ask me one question so we can add a metric."

Sonnet: the daily coach who doesn't get tired of you

Job search isn't one big moment. It's a hundred small tasks every day. Check a letter, prep for a call, reword one bullet, decide whether to bother applying at all. Sonnet is built for this mode.

It holds context within a conversation well, so you can build on things incrementally. And it doesn't overdo it. Where Opus sometimes gives you three screens when you needed two lines, Sonnet keeps things proportional.

Exact prompts for Sonnet

1. 1Prep for recruiter screening call. "Here's the job: [JD]. Here's my CV: [CV]. I have a 30-minute screening call tomorrow. Give me 5 questions they'll probably ask and a short response guide for each."
2. 2Should I apply or not. "Here's the job posting. Here are three areas where I don't fully match: [list]. Is it worth applying at all? What actually decides it in a situation like this?"
3. 3Iterative bullet rewrite. "Here's my bullet: '[text]'. Make it more specific. Then make it shorter. Then give me both versions and one line on the difference between them."
4. 4Salary negotiation breakdown. "I've been offered [amount] for a [title] role in [city/country]. I want to ask for [amount]. Write me a reply email, then give me a short script if they push back." Or head straight to the AI Coach, which has a dedicated negotiation scenario.

Fable: when the letter has to actually land

Honestly, most AI-written cover letters read identically. Recruiters who go through a hundred a day spot the patterns immediately. 'I am excited to apply for...' and they've mentally closed the tab.

Fable was trained to make text sound like something. It has a better feel for sentence rhythm, keeps a voice more consistently, and avoids clichés better. But it needs more input from you, because it won't invent details. Give it details and it'll make something worth reading.

Exact prompts for Fable

1. 1A letter with a voice, not a template. "Here's the job: [JD]. Here are three things I'm genuinely interested in about this company or role: [1, 2, 3]. Here's one specific situation from my experience that's relevant: [short description]. Write a cover letter under 250 words that sounds like a person wrote it, doesn't open with a cliché, and doesn't end with 'looking forward to hearing from you'."
2. 2Unconventional opening paragraph. "Most letters for this job open the same way. Write three different opening paragraphs for my letter, each from a different angle: one through a specific problem I solved, one through something that surprised me about the company, and one that's short and slightly unexpected. Then tell me which one you'd actually open."
3. 3Editing your own draft. "Here's my draft letter: [text]. Keep my voice and my logic, but make it 30% more alive. Don't change the facts. If there's a sentence that sounds particularly templated, replace it and explain why."

How to pick the model in practice

Short version: the task picks the model. No need to overthink it every time.

• CV needs strategic analysis, a fit breakdown against a job description, or there's a complex career situation (direction change, 2022 gap, relocation and market switch at once) - Opus.
• Daily interview prep, rewriting a couple of bullets, working through an offer, deciding whether to apply - Sonnet. Or straight into the AI Coach if you want structured scenarios.
• Cover letter, especially for roles where competition is high and you need to sound like a person rather than everyone else - Fable.
• Quick 'what even is this company' or 'explain this technology to me' - any of them, doesn't matter.

Where Claude won't help (and you should know this)

Honestly. Claude doesn't know your specific company from the inside. It doesn't know what this particular hiring manager is actually looking for right now. It doesn't see the job market in real time, and it especially doesn't know the nuances of the Ukrainian IT market in 2026.

That's why prompts with specific context beat generic ones by a mile. 'Write me a good CV' - bad. 'Here's the JD, here's my CV, here are three things I'm unsure about' - good. The difference in output quality is genuinely large.

One more thing: if you get something from Claude that sounds nice but hollow, don't accept it. Push back. 'This sounds generic, be more specific for a backend dev at Series B/C startups'. The models respond well to pushback, no need to be polite about it.

• Always give context: role, seniority, market, specific job posting if you have one.
• If the answer is generic - specify or ask for 'concrete examples only'.
• Don't use Claude as the final step. Use it for the draft, then edit it yourself.
• If you're writing in English and it's not your first language, ask Fable to check the tone after: 'does this sound non-native? which three spots give it away most?'