Claude Large Language Models

Launch Announcement Prompts: 6 for Separating Claim From Evidence

GPT-6 Astra launched with an executive calling it a generational leap, no published API pricing and no independent benchmarks. That combination is normal and it is also the point. These six prompts are for reading any model launch and working out which parts are measurements, which are projections, and which are marketing. They apply to every vendor equally.

⌨️ 6 prompts 🕐 Updated Sep 4, 2026

💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.

Sort the claims by who can verify them

The last instruction matters. Paraphrasing is where hedged claims become assertions.

Here is a model launch announcement: [PASTE]

Sort every substantive claim into:
VERIFIED — a third party could check this today
VENDOR-MEASURED — a number the company produced itself
PROJECTED — a statement about future capability or interpretation
UNFALSIFIABLE — phrased so no evidence could contradict it

Quote the exact wording for each. Do not paraphrase claims into a stronger or weaker form than they were made in.

Find what the announcement does not say

Omissions are usually more informative than the claims. Pricing absent at launch is a fact worth noticing.

Read this launch announcement: [PASTE]

Tell me what a buyer would need to know that is not here. Specifically:
- Pricing, and whether any is stated
- Availability dates versus rollout language
- What was compared against, and whether it was current
- Any limitation mentioned only in passing

If something important is absent rather than unfavourable, say so plainly.

Check what a benchmark actually measures

Benchmark names sound self-explanatory and rarely are. The last line prevents a confident guess.

A launch cites this benchmark result: [BENCHMARK AND SCORE]

Tell me:
- What the benchmark actually tests
- What a score of that level means in practical terms
- What it does not tell me
- Whether the comparison model was the current version at time of testing

If you do not know the benchmark, say so rather than inferring from the name.

Weigh a claim made against self-interest

A safety disclosure is usually the most credible line in a launch, because it costs the company something. Worth reading closely.

This announcement includes a disclosure that appears to work against the company: [PASTE]

Tell me:
- Why a company might publish this
- Whether it is more or less credible than the promotional claims, and why
- What it implies that is not stated
- What would make me doubt it

Do not assume good faith or bad faith. Reason from incentives.

Work out when to revisit

Most launch decisions should be deferred. This turns that into a plan rather than procrastination.

A new model launched with: [WHAT IS KNOWN] and missing: [WHAT IS NOT]

Tell me:
- What information would change my decision
- Roughly when each usually becomes available after a launch
- What I should be doing in the meantime
- What would make me act sooner

I want a date to revisit, not a recommendation now.

Compare two announcements on the same terms

Vendors choose benchmarks that flatter them. Finding where two announcements are incommensurable is more useful than a verdict.

Here are launch announcements from two vendors: [PASTE BOTH]

Compare them on:
- What each actually committed to versus described
- Which claims are checkable
- Where they use different measures for the same thing
- What each disclosed that the other did not

Do not pick a winner. Tell me where they are not comparable.