{"slug": "google-is-back-gemini-4-argon-leads-our-vision-benchmark", "title": "Google Is Back! Gemini 4 Argon Leads Our Vision Benchmark", "summary": "Google's Gemini 4 Argon returned to the top spot on Roboflow's overall vision benchmark, scoring 82.5% and 85.1% mAP@50 on object detection at low and high effort settings, respectively, and beating GPT-6 Astra at both effort levels. Roboflow reported that Argon improves on the prior best Gemini result by more than 10 percentage points for each setting, and that Argon outperformed Qwen 3.8 Max and GPT-6 Astra on box prompting. The evaluation also covered segmentation and three OCR tasks, with Roboflow noting SAM 3 is much faster but limited by weaker understanding of complex text prompts.", "body_md": "Since we started evaluating vision-language models, Gemini have consistently scored high on our detection and text benchmarks. Recently, however, other research groups have accelerated their progress.\n\nGPT-6 Astra took the lead on most of our benchmarks, trailed by Anthropic and Qwen, which both launched increasingly capable models, including some smaller and quicker variants, resulting in Google slipping in the rankings. With the launch of Gemini 4 Argon, Google returns to the top spot on our overall vision benchmark.\n\nIn this post, we will look at what is possible with Argon across object detection, segmentation, and three OCR tasks. We will also test more advanced video use cases. Along the way, we will discuss the prompts, speed, and cost of running these experiments.\n\nGet starte\n\nExplore Gemini 4 Argon in Roboflow Playground. Try object detection, segmentation, and OCR on your own images.\n\nLet's start with object detection using text prompts. We give the model an image and a list of classes, then ask it to return bounding boxes for the objects belonging to those classes.\n\n```\nDetect all objects in this image belonging to these classes:\n\n\nReturn only a JSON array with this structure:\n[\n {\n \"box_2d\": [y_min, x_min, y_max, x_max],\n \"label\": \"\"\n }\n]\n\nUse only the supplied class names.\n\nCoordinates must be integers normalized to 0–1000,\nrelative to the image height and width.\n\nDo not include explanations or Markdown.\n```\n\nArgon claims the top position on our object detection leaderboard for both low and high effort settings, scoring 82.5% and 85.1% mAP@50, respectively. It outperforms Astra at both effort levels, and improves on the prior best Gemini result by over 10 percentage points for each setting. The model delivers strong results across many object types and varied scenes.\n\nBox Prompting\n\nInstead of telling Argon what you want to detect, you can simply show it. Here, you show the model an example of an object and ask it to find more objects of the same kind. We previously explored this with Qwen 3.8 Max and GPT-6 Astra, and Argon performed best.\n\nYou can give the example boxes as coordinates in the prompt, or highlight them directly on the image. Positive samples indicate what to find, while negative ones clarify what should be ignored. This is very helpful with classes that are tricky to define in words or when text descriptions are ambiguous.\n\n```\nFind objects matching the positive examples and excluding\nthe category represented by the negative examples.\n\nExample boxes:\n{\n \"positive\": [],\n \"negative\": []\n}\n\nExclude the supplied example objects from your predictions.\nInfer the target category without naming it.\n\nReturn only:\n[\n {\n \"box_2d\": [y_min, x_min, y_max, x_max],\n \"label\": \"object\"\n }\n]\n\nAll input and output boxes use integer coordinates normalized\nto 0–1000 relative to the image. Return [] if there are no matches.\n```\n\nYou can provide a reference object in one image and ask the model to find matching objects in another. In the final chess example, we use this approach to detect pawns across images.\n\nSegmentation\n\nSegmentation is a relatively resource-intensive task for VLMs. The model is asked to represent each object as a polygon, meaning it must generate points for all parts of the boundary. Objects with holes or separate parts require extra structure, and outputs can quickly become large.\n\nEven with that limitation, the quality of the masks shows that a lot is possible. The carrots example is particularly interesting because the foreign objects contain holes that need to remain outside the predicted masks.\n\nThe obvious comparison is prompting SAM 3 with text to get masks. SAM 3 is much faster, but its understanding of complex text prompts can be a limitation, especially when the description requires distinguishing between similar objects. Argon's stronger language understanding makes it useful when specifying what to segment takes more than a simple class name.\n\nOCR\n\nSingle String Extraction\n\nOur first text-oriented task involves extracting a single string. Curiously, this is one area where earlier Gemini models remained strong even once Astra had launched. Argon widens that advantage, achieving 80.0% in high effort mode, versus 71.7% for the previous Gemini best at the same effort setting.\n\nThere are still plenty of cases where the model fails. Small, blurred, or difficult-to-read text remains a challenge, as shown by the license plate example. Another source of errors is when reading the image requires reasoning beyond recognizing characters, such as interpreting the individual dials of an electricity meter and combining them into one reading.\n\nJSON Data Extraction\n\nArgon also tops our leaderboard for JSON data extraction, though the gap between leading models is narrower for this task. It reaches 92.6% at low effort and 94.8% at high effort. This is an improvement over previous Gemini models, but those models were already very accurate on this task.\n\nHandwriting continues to be problematic. Examples with handwritten cheques and pharmacy bills illustrate this challenge.\n\nText Localization and Recognition\n\nText localization and recognition is one of the newer tasks in our benchmark. It combines detecting where the text is with reading what it says, so only a handful of models score highly. The latest OpenAI models were ahead of the pack, but Argon now leads at both effort levels.\n\nThis task is useful for processing technical drawings, maps, and other images where text is spread across different locations and orientations. The model needs to return both the text and its location, including when the font is unusual or the characters are rotated. The examples below show technical dimensions, a vintage gasoline sign, and rotated letter tiles.\n\nVideo Reasoning\n\nRe-identification\n\nWe also tested whether Argon could identify the same players across different video frames. We had already performed a similar experiment with Astra and were curious whether we could replicate it with Argon. We gave the model two frames from the same basketball play and asked it to return bounding boxes with consistent player IDs across both images.\n\n```\nIdentify all active basketball players in two chronological frames,\n