Misleading: Jev was developed by the co-inventor of ChatGPT.
Checked by
@insta_factchecker_bot on Telegram
MISLEADING
2 claims: 2 misleading — Jev was released on 15 September 2026 by TypeSafe AI, co-founded by Diogo Almeida.
The claims
Misleading: Jev was developed by the co-inventor of ChatGPT.
Jev was released on 15 September 2026 by TypeSafe AI, co-founded by Diogo Almeida. Almeida was an OpenAI researcher who co-authored the 2022 InstructGPT paper (*Training language models to follow instructions with human feedback*, Ouyang et al., arXiv:2203.02155) and contributed to the RLHF pipeline behind ChatGPT and GPT-4. However, ChatGPT was built by a large institutional team at OpenAI, not an individual "inventor". Labeling himself or being labeled "the co-inventor of ChatGPT" has already drawn Community Notes on X for overstating his role.
Where this comes from: OpenAI paper author lists (arXiv:2203.02155) and TypeSafe AI corporate filings.
Misleading: Jev rebuilt a 586-page internal link map in 45.1 seconds for $0.21, making it ~190x cheaper and far faster than Claude Opus 5.
The raw math of the author's benchmark holds (at 21 pages for $1.43, Opus cost ~$0.068 per page, while Jev cost ~$0.000358 per page, which is ~190x cheaper), but the comparison is an engineered strawman. Jev is a discriminative "System One" model that outputs non-generative classification probabilities in ~100-350 ms at $0.042 per million input tokens. Claude Opus 5 is a heavy, autoregressive frontier LLM running token-by-token generation at roughly $5 input and $25 output per million tokens. Using Opus to execute 8,790 binary yes/no linking decisions is using a flagship reasoning model for a job meant for a classification head, chosen solely to manufacture an extreme cost and latency divergence.
Where this comes from: Author's proprietary SaaS demo ("Internal Link Dealer") via pricing sheets for TypeSafe AI and Anthropic.
Evidence
- arxiv.org · primary
The messenger
@borja.obeso — Borja Obeso is the co-founder of Distribb (distribb.io), an AI-driven SEO automation SaaS ($97-$495/month). He holds no formal AI research credentials. The video is a top-of-funnel lead magnet ("Comment JEV for early access") designed to collect contacts and promote Distribb's upcoming Internal Link Dealer feature. The on-screen dashboard demonstrates real software execution, but against an intentionally inefficient baseline.
Credibility: Mixed — While the performance metrics for his run are mechanically reproducible, the technical claims are heavily filtered through marketing hype to sell software subscriptions.
If you want to reply to the reel
Comparing a discriminative classification model to an autoregressive frontier LLM on binary link decisions is a rigged benchmark designed to manufacture a 190x cost gap.
How this was checked
The reel's video, audio and on-screen text were read directly, and each claim was checked against sources found by live web search. Only sources a reader could verify themselves are cited; social posts and blogs are read but never counted as evidence. The verdict on the page is the worst-rated claim, because that is the part worth knowing before you share.
About this check
Checked 18 Oct 2026 · gemini-3.7-flash · the original reel by @borja.obeso
Think something here is wrong? Reply to the bot with a source and it will be looked at again.
Link to this check
https://reelcheck.pages.dev/p/check-Ddgp5wfSJil