📰 Qwen3.5-Omni Beats Gemini-3.1 Pro in 2026 Multimodal AI Benchmark — Costs 90% Lower Qwen3.5-Omni,
benchmarks gemini huggingface multimodal qwen
| Source: Mastodon | Original article
Alibaba’s Qwen3.5‑Omni has topped Google DeepMind’s Gemini‑3.1 Pro on the 2026 multimodal AI benchmark while slashing input‑token costs to under $0.08 per million—a price‑tag roughly one‑tenth of Gemini’s $2‑per‑million rate. The result, released on March 31, follows the company’s earlier claim that the model “outperforms Gemini” and adds hard data from a suite of vision‑language, audio‑transcription and code‑generation tests.
Qwen3.5‑Omni, built on the 35‑billion‑parameter Qwen3.5‑35B‑A3B architecture and offered as the hosted Qwen3.5‑Flash service, supports a 1 million‑token context window and ships with built‑in tool use. Its open‑source Apache 2.0 licence lets developers run the model locally, while the cloud version bundles production features that were previously exclusive to enterprise‑grade offerings.
The cost advantage matters because multimodal workloads—image captioning, video analysis, real‑time translation—have traditionally been priced out of many Nordic startups and public‑sector projects. By delivering comparable or superior accuracy at a fraction of the expense, Qwen3.5‑Omni could accelerate adoption of AI‑augmented products across fintech, health tech and media in the region. The price gap also pressures Google to justify Gemini‑3.1 Pro’s premium, potentially reshaping the competitive landscape for large‑scale foundation models.
Looking ahead, Alibaba plans to roll out a 397‑billion‑parameter variant that, according to Unsloth documentation, sits in the same performance tier as Gemini‑3 Pro, Claude Opus 4.5 and GPT‑5.2. Observers will watch whether the larger model retains the low‑cost token economics and how cloud providers price the service. Google’s response—whether through price cuts, new feature releases or tighter integration with its own ecosystem—will be the next barometer of market momentum. The coming months should reveal whether Qwen3.5‑Omni can convert its benchmark win into sustained market share.
Sources
Back to AIPULSEN