Claire Vo of ChatPRD on GPT-5's developer-friendly design and the need for role-specific models

Summary

Claire Vo, CEO of ChatPRD, offers one of the more grounded early-user critiques of GPT-5, describing it bluntly as built 'for developers, by developers.' Her assessment carries operational weight: ChatPRD A/B tests every model rollout against metrics including output quality and token generation, and the early GPT-5 numbers are striking.

Token Bloat Is a Real Cost Signal

Vo reports GPT-5 generates four to ten times the tokens of fourth-generation models in her application environment. For a business writing and product strategy platform, that is not a feature. Verbosity inflates inference costs and degrades user experience in contexts where conciseness signals competence. She also flags an aggressive default toward bullet points and tactical, execution-level framing that resists redirection toward higher-order strategic thinking.

The Current Test Matchup

ChatPRD is running GPT-4o-1 against GPT-5 in live testing. GPT-4.5 was Vo's preferred prose model on quality alone, but latency made it commercially unviable. GPT-4o-1 emerged as the user-validated balance of quality, intelligence, and performance. GPT-5 is now the challenger, but Vo says her team needs to rework prompts before it is ready for general rollout, specifically to suppress bullet-point defaults and reduce unnecessary tool calls.

Where GPT-5 Does Win

Technical specification generation is the clear exception. Vo says GPT-5-produced technical specs fed into agentic coding and prototyping frameworks, including v0.dev and Lovable, produce materially higher-quality outputs than prior models. ChatPRD, which recently launched an MCP integration and is accessible inside Cursor, positions itself as the product layer sitting upstream of those engineering execution tools.

The Case for Role-Specific Models

Vo's broader argument is structural. She wants OpenAI to move away from numbered general-purpose releases toward role-tuned models, something analogous to a 'GPT Developer' or 'GPT Strategist,' pre-configured for the context in which they will operate. Her analogy is direct: 'You don't send your engineer into your executive meeting.' As long as flagship models are optimized for software engineering tasks, non-developer applications will absorb friction and tuning overhead that erodes the productivity case.

Platform Bet on OpenAI

Despite the criticism, Vo says ChatPRD remains committed to the OpenAI ecosystem, citing developer experience and API primitives as a genuine competitive differentiator over other model providers. Her asks of OpenAI are specific: better out-of-the-box tooling, finer model control, and expanded hosted services, positioning developer infrastructure, not raw model intelligence, as the next axis of platform competition.