David Ondrej tests GPT 5.5 against Claude Opus models
YouTuber David Ondrej reviews OpenAI's newly released GPT 5.5, testing it in Codex and comparing it to Anthropic's Claude Opus lineup amid claims that Anthropic is facing compute constraints.
Score breakdown
Watch for GPT 5.5's performance in Codex-based coding workflows, as the video frames it as a direct challenger to Claude Opus for front-end and agentic coding tasks at a time when Anthropic's top models are drawing user criticism.
- 01GPT 5.5 was released by OpenAI and made available first to Pro users at the time of filming.
- 02Ondrej tests GPT 5.5 inside the Codex app, which received a UI update around the same time as the model launch.
- 03GPT 5.5 is described as being on the level of Anthropic's unreleased 'Mythos' model, with reported strengths in SVG graphics, 3D, and front-end web development.
David Ondrej's video covers the surprise release of GPT 5.5 by OpenAI, which he describes as the company's direct response to Anthropic's unreleased "Mythos" model. A key distinction he highlights is availability: GPT 5.5 launched immediately for Pro users, while Mythos has not yet been released. He tests GPT 5.5 inside the Codex app — which received a UI update around the same time — and notes the model is reportedly exceptional at SVG graphics, 3D rendering, and front-end web development. He also demonstrates using GPT 5.5 inside ChatGPT with thinking mode enabled to summarize the model's own release page.
A significant portion of the video is devoted to Ondrej's view that Anthropic is struggling with compute constraints.
A significant portion of the video is devoted to Ondrej's view that Anthropic is struggling with compute constraints. He argues that Opus 4.7 is the first Anthropic model in two years that users widely perceive as worse than its predecessor, with regressions spotted in Opus 4.6 and 4.7, degraded performance in Claude Code, and users hitting usage limits quickly. He states he personally defaults to Opus 4.6 (specifically the "4.6 fast" mode in Claude Code) because it follows instructions more reliably than 4.7. By contrast, he contends OpenAI has invested substantially more in compute infrastructure and data centers, which he believes will enable it to produce better models throughout 2026 — framing GPT 5.5 as a potential first signal of that comeback. The transcript is truncated before the full hands-on Codex testing segment concludes.
Key facts
- 01GPT 5.5 was released by OpenAI and made available first to Pro users at the time of filming.
- 02Ondrej tests GPT 5.5 inside the Codex app, which received a UI update around the same time as the model launch.
- 03GPT 5.5 is described as being on the level of Anthropic's unreleased 'Mythos' model, with reported strengths in SVG graphics, 3D, and front-end web development.
- 04Ondrej argues Opus 4.7 is the first Anthropic model in roughly two years that users widely perceive as worse than its predecessor (Opus 4.6).
- 05He states he personally still uses Opus 4.6 ('4.6 fast') in Claude Code daily, citing better instruction-following and the absence of a fast mode in 4.7.
- 06Ondrej attributes Anthropic's model quality issues to compute constraints, contrasting this with OpenAI's heavier investment in data center infrastructure.
- 07