Updates
Site changelog and a release watch for the tools we test. We don’t collect email addresses: to follow along, come back to this page or subscribe to the RSS feed.
When a tool ships a major update, we re-run the affected tests and show before/after.
AI tool release watch
The latest major release of 10 important tools in our lineup, from the vendor’s own page. Each row checked on .
-
Gamma
Latest major release: Gamma 5
Re-test planned: No. Not tested yet; our first presentations test (early November 2026) will run on Gamma 5.
-
Gemini image (Nano Banana)
Latest major release: Nano Banana 2.1
Re-test planned: No. Not tested yet; our first image test (second half of October 2026) will record exactly which Nano Banana model it ran on.
Source: Gemini API release notes (opens in a new tab) Tool page
-
Kling AI
Latest major release: Kling 4.0 (early access; 4.0 Flash for some subscribers)
Re-test planned: Yes. Kling 4.0 is in early access, with a wider rollout announced for October 2026. If it’s publicly available before our video round (late October), we test 4.0. If it arrives later, we re-run the video tests on 4.0 and show before/after.
-
ElevenLabs
Latest major release: Eleven v4 and Eleven v4 Turbo
Re-test planned: No. Not tested yet; our first voice test (early November 2026) will run on Eleven v4.
-
ChatGPT Images (OpenAI)
Latest major release: ChatGPT Images 2.5
Re-test planned: No. Not tested yet; our first image test (second half of October 2026) will run on ChatGPT Images 2.5.
-
Midjourney
Latest major release: V8.2 (default model)
Re-test planned: No. Not tested yet; our first image test (second half of October 2026) will run on V8.2.
-
Synthesia
Latest major release: Express-3 avatar model
Re-test planned: No. Not tested yet; our first avatar-video test (late October 2026) will run on Express-3.
-
HeyGen
Latest major release: Avatar V avatar model
Re-test planned: No. Not tested yet; our first avatar-video test (late October 2026) will run on Avatar V.
-
Google Veo (Flow)
Latest major release: Veo 3.1 Lite (a cheaper tier of Veo 3.1)
Re-test planned: No. Not tested yet; our first video round (late October 2026) will record exactly which Veo model it ran on.
Source: Gemini API release notes (opens in a new tab) Tool page
-
Runway
Latest major release: Gen-4.5
Re-test planned: No. Not tested yet; our first video round (late October 2026) will run on Gen-4.5.
Site changelog
-
Lab run #1: first real results (Workers AI models)
We published our first archived lab run: exact-text image prompts and a synthetic 24-minute meeting transcript, scored with OCR + visual check and Whisper-normaliser WER. Scope is the Cloudflare Workers AI models we can reach on our own account — not consumer apps. Consumer tools stay NOT YET TESTED. Full run: /lab/runs/lab1-20261008T2035Z/
-
Where testing stands
Methodology v0.2 is published. First archived lab results are live (Workers AI models only — see Lab run #1). Consumer apps tested: 0. Broader consumer-tool results are still planned for the second half of October 2026 (meeting notes and text in images), then video (late October), then voice and presentations (early November).
-
New: this updates page and an RSS feed
We removed the re-test alert form. It never stored an email address, so it shouldn’t have asked for one. New results, re-tests and major tool releases are posted here and in the RSS feed instead.
-
Honest labels while nothing is tested
Ranking, category and comparison pages now say that testing is in progress and when first results are planned, instead of “ranked”. Until at least one tool on a page is tested, the page is hidden from search engines.
-
Comparisons only within one job
We removed comparisons between tools that do different jobs (for example an image suite vs an avatar-video tool). Comparison pages are now generated only for tools built for the same job.
-
A clearer home page
The home page now says in one line what to do, with three steps and clickable examples. Answers have a “Try [tool]” button that opens the tool directly. Where we haven’t tested yet, the answer says so and shows the options instead of picking a winner.