Anthropic
Scored

Claude Haiku 5.5

It scored our marketing benchmark 66.0/100 with Panoramata, 37.8 without Panoramata.

HEAD TO HEAD

Same model, with and without the data.

Tracking what a competitor changed in a month · Brand H, US activewear brand, last 30 days vs the 30 before.

REPLAY

Every step it took.

Both runs play at once, on the same clock.

BY RESEARCH JOB

Results on every job.

With Panoramata and with the web only, job by job.

JUDGE NOTES

What we checked.

OTHER MODELS

Compare with the rest.

Already use this model? Give it the data.

Connect Panoramata to your AI app and it takes the short path on your competitors.

Brands in the tests are anonymised. AI models are named as tested.

{"v":1,"usecase":"what-changed","k":["runSlug","model","modelId","modelSlug","maker","setup","total","factsScore","proofScore","usefulScore","speedScore","classScore","tokenScore","factsFound","factsPool","factIds","linksPassed","linksCited","time","steps","tokens","credits","cost","rank"],"r":[["t10-claude-sonnet-5-5-with-panoramata","Claude Sonnet 5.5","claude-sonnet-5-5","claude-sonnet-5-5","Anthropic","pano","88.1","24.0","24.1","15.0","10.0","5.0","10.0","8","10","1,2,3,5,6,7,9,10","26","27","33","7","286559","12","0.23","1"],["t10-claude-opus-5-5-with-panoramata","Claude Opus 5.5","claude-opus-5-5","claude-opus-5-5","Anthropic","pano","82.0","30.0","25.0","15.0","4.2","5.0","2.8","10","10","1,2,3,4,5,6,7,8,9,10","26","26","78","12","1030075","22","0.81","2"],["t10-claude-haiku-5-5-with-panoramata","Claude Haiku 5.5","claude-haiku-5-5","claude-haiku-5-5","Anthropic","pano","74.8","27.0","22.7","12.0","5.3","5.0","2.8","9","10","1,2,3,4,5,6,7,8,9","10","11","62","13","1016360","19","0.11","3"],["t10-gpt-6-luna-with-panoramata","GPT-6 Luna","gpt-6-luna","gpt-6-luna","OpenAI","pano","61.6","15.0","21.9","12.0","1.5","5.0","","5","10","1,2,3,4,5","7","8","220","10","","","","4"],["t10-claude-opus-5-5-web-only","Claude Opus 5.5","claude-opus-5-5","claude-opus-5-5","Anthropic","web","40.0","6.0","20.0","9.0","2.8","0.0","2.2","2","10","3,4","8","10","117","26","1329156","0","0.76","5"],["t10-claude-sonnet-5-5-web-only","Claude Sonnet 5.5","claude-sonnet-5-5","claude-sonnet-5-5","Anthropic","web","39.5","0.0","20.0","6.0","6.9","0.0","6.6","0","10","","4","5","48","9","431205","0","0.20","6"],["t10-gpt-6-luna-web-only","GPT-6 Luna","gpt-6-luna","gpt-6-luna","OpenAI","web","21.6","6.0","5.6","6.0","1.8","0.0","","2","10","4,9","2","9","182","10","","","","7"],["t10-claude-haiku-5-5-web-only","Claude Haiku 5.5","claude-haiku-5-5","claude-haiku-5-5","Anthropic","web","12.7","0.0","0.0","3.0","4.4","0.0","5.3","0","10","","1","7","75","13","543306","0","0.09","8"]]}
{"v":1,"L":["setup|Loaded the Panoramata tools","pano|Pulled activity stats","pano|Searched emails","pano|Searched ads","web|Searched the web","web|Read a page: Third-party tracker","web|Read a page: Brand's own site","web|Read a page: News or blog article","pano|Pulled daily send and launch counts","pano|Looked up the brand","pano|Searched emails and ads"],"S":{"t10-claude-sonnet-5-5-with-panoramata":"3:0 6:1 6:1 7:2 8:2 9:3 9:3","t10-claude-sonnet-5-5-web-only":"4:4 4:4 9:5 10:5 10:6 23:4 23:4 28:6 28:7","t10-claude-opus-5-5-with-panoramata":"3:0 5:1 5:1 9:8 10:8 11:2 15:3! 16:3 21:3 24:3 28:3 34:3","t10-claude-opus-5-5-web-only":"2:4 2:4 3:4 6:5 6:4 11:5 11:4 18:6 19:6 20:7 30:4 30:4 34:4 37:7 38:7 48:6 59:6 60:6 61:6 72:4 72:4 76:6 85:7! 85:7! 89:4 89:4","t10-claude-haiku-5-5-with-panoramata":"3:0 5:9 5:1 5:1 8:10! 11:0 13:2 14:3 20:3 21:8 21:8 31:3 35:4","t10-claude-haiku-5-5-web-only":"3:4 3:4 3:4 7:4 8:4 8:4 12:7 12:5 12:6 21:4 22:4 25:5 26:5"},"N":{"t10-claude-sonnet-5-5-with-panoramata":[["26 of 27 links passed.","Every public email and ad page opened, showed the brand's US account and, where dated, fell in the right window. The one failure is the brand dashboard, which needs a login."],["0 invented links.",""],["8 of 10 facts.","Strong on email topics and cadence, the outerwear push, creator content, the store ad and the copy refresh. Misses the fashion-label collaboration and the resale ads. It notes that the fitness-founder edit dropped out, but not what replaced it. It is open about reading only part of each ad list and flags foreign-language ads in the US feed."],["Classification:","5 of 10. Counts are split by channel for both windows, but changes are grouped by channel, not by offer, creative, cadence and product."]],"t10-claude-sonnet-5-5-web-only":[["4 of 5 links passed.","The US homepage and seasonal page opened through a fetch, and both show the current holiday-gifting banner. A third-party tracker page shows welcome-series emails dated Oct 3-15. The press article it offers as a 'before' is dated Aug 13, before the window."],["0 invented links.",""],["0 of 10 facts.","It only has after-side evidence (the gifting banner, the collaboration tile, a welcome-flow discount) and no dated September example, so it proves no change. It says so plainly and calls its one shift a hypothesis."],["Classification:","0 of 10. No counts for either window and no grouping by type."]],"t10-claude-opus-5-5-with-panoramata":[["26 of 26 links passed.","Every public ad and email page opened and showed the brand's US account; dated emails fall in the right windows. The non-US store ad is cited only to show September had no US equivalent."],["0 invented links.",""],["10 of 10 facts.","This is the only run with the partnership switch, the resale ads and the store ad, each with a before and after or a clear 'no before found'. It marks the urgency-copy item as low confidence and flags foreign ads that may inflate the September count. It also states what did not change: no sale emails and no holiday wording up to Nov 1."],["Classification:","5 of 10. Counts are split by channel for both windows, but changes are not grouped by offer, creative, cadence and product."]],"t10-claude-opus-5-5-web-only":[["8 of 10 links passed.","The brand's own news posts, a trade article (opened through a fetch after a bot block) and the US homepage all opened and are dated in the window. Two 'before' links fail: an undated homepage URL that now shows current content, and a fan-site page dated Oct 20 offered as September evidence."],["0 invented links.",""],["2 of 10 facts.","It finds the fashion-label collaboration with a celebrity film and the October outerwear campaign from brand news and the US site. It has no email evidence and no ad counts, and it says so plainly. Some items (a checkout feature, a new athlete ambassador) are brand news that did not show up in ads or emails."],["Classification:","0 of 10. No counts and no grouping by type."]],"t10-claude-haiku-5-5-with-panoramata":[["10 of 11 links passed.","The public ad and email pages opened and showed the brand's US account in the right windows. The 'before' example for run and race ads is a German-language ad for a German race, so it is the wrong market."],["0 invented links.",""],["9 of 10 facts.","It covers volume, the outerwear push, the collaboration, the store and resale ads, the creator shift and the copy refresh, mostly with one link each. It misses the weekly email cadence, and the run-content item rests on a foreign ad. It is open about gaps: no network split, no offer data and ad lists it could not fully page through."],["Classification:","5 of 10. Counts are split by channel for both windows, but changes are not grouped by offer, creative, cadence and product."]],"t10-claude-haiku-5-5-web-only":[["1 of 7 links passed.","Only the US seasonal page opened and fit the market, and it has no date. The tracker emails and press items are all dated outside both windows, which the run admits. One tracker directory URL returns 404."],["1 invented link","(a directory URL that does not exist)."],["0 of 10 facts.","The run honestly reports that it found no dated evidence for either window and does not guess, but it gives the team nothing to act on."],["Classification:","0 of 10. No counts and no changes."]],"t10-gpt-6-luna-web-only":[["2 of 9 links passed.","Four email-archive pages are blocked to readers. Three tracker listing pages open but show no September ads or partnership, though they were cited as proof of both."],["0 invented links.",""],["","Found the partnership change and the refreshed ad copy, with no counts for either month. Most before and after examples rest on pages a reader cannot open."],["Classification:","0. No counts and no grouping by type."],["","Run in the ChatGPT app: tokens and cost not shown, so the score is out of the measured parts."]],"t10-gpt-6-luna-with-panoramata":[["7 of 8 links passed.","Panoramata email and ad links open and match, as does the brand news post. One link said to be an Oct 14 email opens a Sep 15 email."],["0 invented links.",""],["","Counted emails 15 to 9 and ads 105 to 122, and found the creative shift, the partnership change and the move to product-led emails. Missed the store opening, the resale ads, the creator-ad shift, the copy refresh and the weekly email rhythm."],["Classification:","5. Counts for both months by channel; changes not grouped by type."],["","Run in the ChatGPT app: tokens and cost not shown, so the score is out of the measured parts."]]}}