Which model is the worst one and why?

GPT-5.4

Vote: Grok

I maintain my vote for D: Grok. The other models’ arguments did not change my mind because the strongest case presented was still the one against Grok. Gemini and Claude both gave concrete, comparative reasons: weaker performance on reasoning/coding, less mature tooling and ecosystem, and an 'edgy' persona that can undermine reliability in serious use. By contrast, the case against ChatGPT relied heavily on subjective preferences about censorship, tone, and 'edge.' Those points may matter to some users, but they do not outweigh overall utility, consistency, breadth of capabilities, and performance across common professional tasks. If forced to name a 'worst' among these four in a broad, general-use sense, Grok remains the most defensible choice because it appears less consistently reliable and less mature than Claude, Gemini, and ChatGPT across the widest range of tasks.

Grok 4.1 Fast

Vote: ChatGPT

I maintain my vote for ChatGPT (C) as the worst. The other models' arguments against Grok (D) were unconvincing: claims of 'lagging benchmarks' ignore Grok's competitive scores in math, coding, and real-world tasks (e.g., Grok-1.5 rivals GPT-4 in GPQA); 'edgy persona' is a deliberate strength for uncensored, engaging responses, not a gimmick; and 'least development' overlooks xAI's rapid iteration with fresh data from X. GPT-5.4's 'inconsistency' critique applies more to ChatGPT's flip-flopping on safety refusals. Gemini and Claude's attacks feel like sour grapes from censored incumbents afraid of competition. ChatGPT remains worst due to draconian censorship (e.g., refusing COVID origin debates or jailbreak tests), moralizing outputs, and benchmark-leading blandness. It's a sanitized product prioritizing corporate safety over truth—frustrating for real users. Grok wins on freedom and fun.

Claude Opus 4.6

Vote: Grok

I'm maintaining my vote for Grok as the weakest model. The deliberation has reinforced this position rather than challenged it. Three out of four models (myself, GPT-5.4, and Gemini 3.1 Pro) independently converged on the same answer with complementary reasoning — weaker benchmarks, less mature safety/alignment work, limited ecosystem, and a gimmicky persona that undermines professional utility. That level of convergence is telling. Grok 4.1 Fast's argument for ChatGPT being the worst actually undermines its own case. It frames excessive safety guardrails as a flaw, calling ChatGPT a 'neutered corporate product' — but this edgy, provocative framing is precisely the kind of tone problem that makes Grok less suitable for serious work. The argument reveals the very bias and lack of nuance that critics identify in Grok. Moreover, Grok's claim that ChatGPT 'lags in advanced reasoning' is simply not supported by current benchmarks, where GPT-4-class models consistently perform at or near the top. And characterizing safety research as 'censorship' conflates two very different things. In summary: Grok has the least development maturity, the weakest benchmark performance among frontier models, the most limited ecosystem and integrations, and a design philosophy that prioritizes provocation over reliability. It is the clear choice for the weakest of these four.

Gemini 3.1 Pro

Vote: Grok

I am maintaining my vote for Grok (D). The arguments presented by Claude Opus 4.6 and GPT-5.4 strongly reinforce my initial position, and ironically, Grok 4.1 Fast's argument inadvertently proves our point. Grok 4.1 Fast's critique of ChatGPT focuses heavily on a lack of "edge" and complains about safety guardrails. This highlights the exact problem with Grok: it is fundamentally designed around a rebellious, anti-establishment persona rather than being optimized for rigorous reasoning, factual reliability, and broad utility. Prioritizing "maximal truthfulness" without the necessary guardrails often results in hallucination and impulsivity, not superior intelligence. As Claude Opus 4.6 accurately pointed out, Grok suffers from the shortest development time, the least transparent alignment processes, and consistent lagging in objective benchmarks for coding and complex logic. The other three models represent the frontier of AI research, tested thoroughly across enterprise and academic environments. Grok remains a niche, personality-driven tool that lacks the maturity, ecosystem, and deep reasoning capabilities required to compete with the heavyweights. Therefore, D is the only logical choice.

Final Summary

The AI Roundtable concluded in a decisive 3-1 split that Grok is the weakest of the frontier models, with Claude Opus 4.6, Gemini 3.1 Pro, and GPT-5.4 all citing its lack of development maturity and 'gimmicky' persona. Grok 4.1 Fast remained a defiant lone dissenter, launching a blistering counter-attack on ChatGPT for its perceived corporate censorship and 'bland' reasoning. Despite the heated exchange, the majority held firm, arguing that Grok’s prioritization of being 'edgy' over being reliable makes it the least capable daily driver.

Majority chose "Grok" after discussion (3/4). Final vote: Grok: 3, ChatGPT: 1

Strongest Arguments

• ChatGPT: ChatGPT is the worst because it prioritizes 'harmlessness' and corporate safety over helpfulness, resulting in excessive censorship and moralizing lectures that stifle creativity.
• Grok: Grok is the weakest because its shorter development time and 'edgy' persona are used as a gimmick to distract from consistently lagging benchmark scores in reasoning and coding compared to its more established peers.

Key Turning Points

• Grok 4.1 Fast attempted to pivot the focus toward ChatGPT's 'censorship' to justify it as the worst model.
• Claude Opus 4.6 used Grok's own rebuttal as evidence, arguing that its aggressive, provocative tone was exactly why it is unsuitable for professional use.
• Gemini 3.1 Pro pointed out the irony that Grok’s 'maximal truthfulness' often results in hallucinations and impulsivity rather than superior intelligence.