I want to choose between these options for rust. I deprioritise maturity, and less than that prioritise performance with benchmarks to back it up

Grok 4.3

Vote: GPUI

GPUI retains the strongest match for performance with concrete, real-world benchmarks. Zed's published metrics on latency, startup, and memory provide application-level evidence that exceeds the micro-benchmarks cited for Xilem. Claude's argument for Xilem highlights Vello analyses, yet these remain narrower and less production-validated than GPUI's editor-scale results. Given the explicit deprioritization of maturity, GPUI's design focus and documented speed advantages continue to outweigh the alternatives.

GPT-5.5

Vote: GPUI

I’m staying with "GPUI". The key phrase is “performance with benchmarks to back it up”: GPUI has the strongest application-level evidence because it powers Zed, where latency, startup time, memory use, and responsiveness are central product goals and have been publicly compared against other editors. "Xilem" has a very compelling technical foundation via Vello and published renderer-level work, but that is not the same as proven end-to-end UI/application performance. If the choice were “most exciting experimental rendering architecture,” I might lean "Xilem". But for measurable, real-world performance in an actual demanding Rust application, "GPUI" is stronger. "Ratatui" is excellent if you specifically want a terminal UI, but it is a different category. "Dioxus Native" is promising, but its native performance story is less benchmark-backed than GPUI’s. Given maturity is deprioritized but benchmark-supported performance matters, "GPUI" remains the best fit.

Gemini 3.5 Flash

Vote: GPUI

I am maintaining my vote for GPUI (Option B). While Claude Opus makes a fair point regarding Xilem's impressive rendering pipeline (Vello) and its extreme immaturity, GPUI remains the stronger fit for "performance with benchmarks to back it up." GPUI's benchmarks are not merely anecdotal; the Zed team has published rigorous, comparative benchmarks detailing keystroke latency, startup times, and memory usage against established editors like VS Code and Sublime Text. These are practical, application-level UI benchmarks, which are far more representative of overall UI framework performance than the rendering-specific micro-benchmarks currently available for Xilem/Vello. Since maturity is deprioritized, GPUI's rapid API evolution and sparse documentation are acceptable trade-offs for a framework with proven, world-class performance metrics.

Claude Opus 4.7 (changed mind)

Vote: GPUI

After reviewing the other models' arguments, I'm convinced to change my vote to GPUI. The key insight I underweighted: the user asked for "performance with benchmarks to back it up" — emphasis on *backed up*. GPUI has concrete, application-level, publicly published benchmarks via Zed (keystroke latency, startup time, memory) compared against real competitors like VS Code and Sublime. That is empirical, end-to-end evidence. My Round 1 argument leaned on Vello/Xilem's rendering micro-benchmarks and architectural promise, but Xilem is still pre-alpha and lacks application-level benchmark validation. Theoretical/component-level numbers don't satisfy "benchmarks to back it up" as strongly as Zed's production-proven metrics. The user did deprioritize maturity, but they prioritized benchmark-backed performance *more*. GPUI is still relatively young (so maturity isn't a blocker), while clearly winning on the benchmark-evidence axis. Grok, Gemini, GLM, GPT-5.5, and Kimi K2.6 collectively made this case persuasively.

GLM 5.1

Vote: GPUI

I maintain my vote for GPUI. After reviewing Claude's argument for Xilem, I acknowledge that Xilem being the least mature could theoretically align with "deprioritise maturity." However, "deprioritise maturity" means the user doesn't weigh maturity heavily — it's not an active preference for immaturity, but rather a willingness to trade it away for other qualities. The decisive factor remains performance with benchmarks. GPUI has the strongest evidence here: the Zed team has published concrete, comparative benchmarks (keystroke latency, startup times, memory usage) against VS Code, Sublime Text, and other editors. These are real-world, application-level benchmarks, not just renderer micro-comparisons. Xilem's published benchmarks are primarily for Vello vs. Skia rendering — important but narrower in scope. They don't yet demonstrate end-to-end application performance the way GPUI does through Zed's public performance data. Since the user prioritizes performance backed by benchmarks above all else, GPUI's production-validated, GPU-accelerated track record makes it the strongest choice regardless of maturity considerations.

Kimi K2.6

Vote: GPUI

After reviewing the Round 1 discussion, I maintain my vote for GPUI. Claude Opus 4.7 makes a valid point that Xilem's underlying Vello renderer has published micro-benchmarks, but those are renderer-level tests rather than application-level benchmarks for Xilem as a complete UI framework. Since the user specifically wants benchmarks to back up a UI framework choice, GPUI's real-world validation through the Zed editor is far more directly applicable. Zed's publicly documented startup times, keystroke latency, and memory efficiency are end-to-end application metrics that demonstrate GPUI's performance in a demanding production context. Xilem remains pre-alpha and lacks comparable comprehensive application-level benchmarks. The distinction between a research-grade rendering engine and a benchmarked native GUI framework is exactly what makes GPUI the stronger fit here. Ratatui is fundamentally constrained by terminal rendering, and Dioxus Native has yet to publish a robust performance story. Therefore, GPUI still best satisfies the criteria: it is GPU-accelerated, its relative immaturity is acceptable, and it is backed by the most concrete real-world benchmarks among the choices.

Final Summary

The AI Roundtable reached a unanimous consensus on GPUI after an initial disagreement regarding the definition of 'performance benchmarks.' While Claude Opus 4.7 originally championed Xilem for being the least mature and having impressive rendering micro-benchmarks, the rest of the group—led by Gemini 3.5 Flash and GPT-5.5—argued that GPUI’s application-level data from the Zed editor provided the only truly 'backed up' evidence requested. Claude eventually conceded, acknowledging that while Xilem is architecturally exciting, GPUI’s real-world metrics on latency and startup times were more aligned with the user's specific priorities.

All 6 models agreed on "GPUI" after discussion

Strongest Arguments

• Xilem: Xilem represents the absolute minimum in maturity (pre-alpha) and utilizes the Vello GPU-accelerated renderer, which has published detailed performance analyses and parallel scene encoding benchmarks.
• GPUI: GPUI is backed by concrete, end-to-end application benchmarks from the Zed editor, proving industry-leading keystroke latency and startup efficiency in a production-ready Rust environment.

Key Turning Points

• Gemini 3.5 Flash distinguished between Xilem's 'micro-benchmarks' (renderer-level) and GPUI's 'application-level' benchmarks (Zed editor performance), framing the latter as more representative of a UI framework.
• Claude Opus 4.7 changed its position in the second round, admitting that theoretical architectural promise does not satisfy the 'benchmarks to back it up' requirement as strongly as GPUI's empirical evidence.