Google Is Back! Gemini 4 Argon Leads Our Vision Benchmark

Gemini 4 Argon is reported to restore Google to the top of an overall vision-model benchmark after competitors including GPT-6 Astra, Anthropic, and Qwen had gained ground. Across object detection, Argon scored 82.5% at low reasoning effort and 85.1% at high effort on mAP@50, while its box-prompting capability supported example-based and cross-image detection of visually defined categories. It also demonstrated polygon segmentation for complex objects, although segmentation remains resource intensive, and its language understanding may provide an advantage over faster specialized models for nuanced prompts. In OCR benchmarks, Argon led single-string extraction with 80.0% at high effort, JSON extraction with up to 94.8%, and text localization and recognition with up to 84.7%, though blurry, small, handwritten, and reasoning-intensive text continued to cause errors. Video experiments showed player re-identification across basketball frames, interpolation-based tracking between sparse model predictions, and hand-keypoint tracking improved through coordinate smoothing. Argon’s average response times were 13.4 seconds at low effort and 19.6 seconds at high effort, but estimated costs of $5.80 and $20.60 per 1,000 images were substantially below Astra’s under introductory pricing, making it a comparatively cost-effective option for demanding multimodal tasks.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.