Grok 4.7 vs GPT-6 Astra for Cover Letters: Which One Writes a Letter a Recruiter Reads? (2026)

Two September launches, one job-seeker question

September 2026 gave job seekers two new frontier models inside three weeks. OpenAI released GPT-6 Astra on September 3 (general availability the next day), and it became ChatGPT's default engine on September 25. xAI released Grok 4.7 on September 21 at the same price as Grok 4.6. Both companies pitch their model at 'professional work', and both cite document writing as a strength.

That's the marketing. The question you care about is narrower: when you paste a job ad and your CV and ask for a cover letter, which one produces something a recruiter reads to the end without spotting that a model wrote it?

The answer depends on what each model was built to do. Below are the five things that decide whether a cover letter helps or hurts, and how each model's documented strengths and known failure modes play out on each.

Tone: which letter sounds like a person

GPT-6 Astra inherits the ChatGPT house style, and OpenAI trained it hard on professional-writing benchmarks. Expect a smooth first draft: a specific opener about the company, two achievements, a confident close. That's the compliment and the problem. Recruiters have read a thousand of these since ChatGPT arrived, and the rhythm is recognisable.

Grok 4.7's writing is typically blunter: shorter sentences, fewer transitional phrases, and the occasional line that lands slightly off. The upside is that it's less predictable, and it takes a voice instruction ('write like I talk, not like a template') well.

Whichever you use, give it a sample of your own writing (an email you sent, a LinkedIn post) and tell it to match that. That single input does more for tone than choosing between the two models.

Edge on tone: Astra for a first draft you'll edit, Grok 4.7 if you want it to sound less machine-made out of the box.

Honesty: who invents less

Both models will invent a metric if you give them nothing to work with. That isn't a bug in one of them; it's how language models fill a gap. The difference is what happens when you give them a fact sheet and then ask them to check themselves.

xAI's stated focus for Grok 4.7 was self-verification: the model was trained to re-read its own output and flag what it isn't sure about. That maps directly onto the cover-letter failure mode ('reduced onboarding time by 30%' appearing from nowhere). Ask it to list every unsupported claim and it's built to do exactly that.

Astra is more capable overall (it leads Grok 4.7 on most public intelligence indexes) but OpenAI's release notes emphasise computer use, browsing and professional tasks over self-audit. It will catch its own stretches when asked, but you have to ask, and you may have to ask twice.

The prompt fix is the same for both: end with 'List every claim in this letter that isn't supported by my CV or fact sheet.' Then read every number yourself.

Edge on honesty: Grok 4.7, because verification is what it was tuned for.

Job-ad matching: reading between the lines

A cover letter earns its place by answering the question the job ad is really asking. Both models pull the must-haves out of a posting well. Astra's stronger general reasoning gives it the edge at inferring the unstated priority: if an ad mentions 'ambiguity' three times, Astra is more likely to tell you to lead with a story about shipping without a spec. Grok 4.7 tends to stay closer to the literal text of the ad, which is safer but less persuasive.

For a senior role where the ad is vague and the fit is about judgment, Astra's inference is worth more. For a junior or mid-level role with a checklist of tools, literal matching is fine and less likely to overreach.

Edge on matching: Astra for senior and ambiguous roles, a tie for checklist roles.

Length and format control

Ask either model for 'a cover letter' with no constraint and you'll get 350–450 words. That's too long. Recruiters skim; three short paragraphs at 180–250 words is the target.

Give both a hard cap ('maximum 220 words, three paragraphs') and check the count. xAI specifically called out document and presentation creation as an improvement in 4.7, and hard constraints like word counts are part of that; if a model overshoots, say 'cut to 220' and it will.

Neither model knows the format the employer wants. If the application form has a 'cover letter' text box, it needs plain text with no headers. If it's an attachment, it needs your name and contact line at the top. Tell the model which one before it writes.

Edge on format control: slight edge to Grok 4.7 given its release focus; verify the count on both.

Cost and access

For most job seekers this comes down to what you already pay for. If you have ChatGPT Plus, Astra is the default model as of September 25 and you're already using it. If you have a Grok subscription or use Cursor, you have 4.7.

Through the API, Grok 4.7 lists at $2 per million input tokens and $6 per million output, unchanged from 4.6. Astra is priced at roughly 2.5 times GPT-5.6 Sol. A cover letter is a few thousand tokens either way, so the per-letter difference is cents; it only matters if you're batch-generating letters for 50 applications, which you shouldn't be.

Edge on cost: Grok 4.7 on paper, irrelevant in practice for one letter at a time.

The prompt that works on both

Paste this after your CV, the job ad and a fact sheet of real numbers:

"Write a cover letter for this role. Rules: maximum 220 words in three paragraphs. Paragraph 1: why this company and this role, referencing one specific thing from the ad. Paragraph 2: two results from my CV that map to the ad's top two must-haves, using only numbers from my fact sheet. Paragraph 3: one sentence on what I'd do in the first 90 days, then a plain close. Write in my voice: direct, no 'passionate', no 'I am writing to express'. Then list every claim in the letter that isn't supported by something I gave you."

Run it on both models. Keep the one that sounds like you. Then read every number.

One more thing neither model checks: the CV the letter is attached to. A great cover letter on a CV that an ATS can't parse never reaches a human. Run the CV through a free analysis at https://www.hrlens.io/cv first (score out of 100, five category scores including ATS compatibility, layout check) and paste the same job description so the analysis lists the skills the ad wants that your CV still doesn't show. If you're on All-Access, the CV-aware cover letter on the results page starts from that analysis rather than from a blank prompt, and the standalone writer at https://www.hrlens.io/cover-letter works from role, company and industry when you don't have the CV to hand.

Frequently asked questions