TechCentral launches AI Barometer, an evidence-based, periodically updated snapshot of which AI models perform best for common use cases, drawing on public benchmarks and independent leaderboards.

In everyday assistant and chat tasks, Claude Fable 5 leads the Arena text leaderboard with an Elo around 1,509, followed by Google Gemini 3.1 Pro Preview at 1,486 and OpenAI GPT-5.5 at 1,481, with token costs varying notably by model.

In writing benchmarks, Claude Fable 5 leads EQ-Bench Creative Writing v3 at an Elo near 2,192, narrowly ahead of Claude Opus 4.7 at 2,179, while GPT-5.5 trails at 2,019; a note that the judge is an AI.

Caveats: many results are vendor-reported or statistically tied and not independently verified; results can shift quickly with new releases, with video and music categories moving fastest.

Methodology: The Barometer uses arena-style voting and task-based benchmarking to evaluate models, acknowledging strengths and weaknesses and the possibility of close ties.

Web research: BrowseComp shows GPT-5.5 Pro leading lab-reported scores at 90.1%, with a distinction made between lab-board figures and self-reported numbers.

The project highlights that the Barometer aggregates public benchmarks to reduce vendor marketing bias and provide a transparent performance snapshot.

In music generation, Suno v5.5 tops instrumental and vocal Elo scores around 1,194 and 1,168 respectively, followed by Mureka V8.

Video generation’s arena ranks ByteDance Dreamina Seedance 2.0 first at 1,224, with Alibaba Wan 2.7 at 1,160 and HappyHorse 1.1 at 1,154; Western entrants like Google Veo 3.1 lag at 1,095, indicating a Chinese-led field.

Deep reasoning shows Gemini 3.1 Pro Preview leading Humanity’s Last Exam at 46.4%, with GPT-5.4 Pro at 44.3%; Claude Fable 5’s 53.3% is vendor-reported and not on the official board.

Coding benchmarks place SWE-bench Pro with GPT-5.4 at 59.1% and Muse Spark (Meta) at 55%, with Claude Opus 4.6 at 51.9%; caveat remains that Claude Fable 5’s vendor-reported 80.3% is not independently verified.

Image generation: GPT Image 2 leads with an Elo of 1,339, Reve 2.0 close behind at 1,281, MAI-Image-2.5 in third, with most judgments based on blind human preference.