Sandboxed Coding Agents are Competitive Omni-m: leaderboard
As multimodal LLMs increasingly target video and audio, it is often assumed that such tasks require native omnimodal models. We show that this is not always the case: coding agents with only text+image access and a sandboxed tool-use interface can match, and in several settings outperform, SOTA native omnimodal models and predefined multimodal agent scaffolds across multiple audio-video benchmarks
Metric: OmniGAIA Avg. (self-reported). Source: benchmarklist.com. Saturation forecast: Around July 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (xHigh) | 75 | #79 (GPT-5.4) |
| 2 | Claude Opus 4.6 | 68.6 | #63 |
| 3 | Gemini 3.1 Pro (Preview) | 66.1 | #58 |
| 4 | GPT-5.4 (High) | 64.4 | #79 (GPT-5.4) |
| 5 | Claude Sonnet 4.6 | 63.3 | #90 |
| 6 | GPT-5.4 (Medium) | 56.4 | #79 (GPT-5.4) |
| 7 | Kimi K2.5 | 55.6 | #144 |
| 8 | GPT-5.4 (Low) | 55 | #79 (GPT-5.4) |
| 9 | Gemini 3 Flash | 51.7 | #99 |
| 10 | Qwen 3.5 Omni Plus | 38.7 | #84 |
| 11 | MiniMax-M2.7 | 33.3 | #258 |
| 12 | MiMo-V2-Omni | 25.8 | #253 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sandboxed-coding-agents-are-competitive-omni-m · How It Works · Data refreshed daily, snapshot 2026-10-20.