The Vision AI Playground: test and compare models on your own images

GPT-6.1 Terra is an OpenAI model released on October 26, 2026 for visual reasoning, computer use and writing. It accepts text and image inputs and produces text, with a 1,050,000-token context window and up to 128,000 output tokens. OpenAI lists applications including matching product photographs to catalog entries, interpreting technical drawings and operating application screens.The API exposes six reasoning effort settings: none, low, medium, high, xhigh and max. Medium is the default. The model supports structured outputs and tool calling through the Responses API. Vision Evals reports its Low and High results for text understanding, counting, identification, reasoning and object detection, using native OpenAI API inference. Terra is a proprietary model; its weights are not released.

Muse Spark Open is an open-weight multimodal model released by Meta on October 15, 2026. Its sparse mixture-of-experts architecture has 816 billion total parameters and activates 42 billion per token. It accepts text, images, audio and video, produces text, and supports a context window of 1,048,576 tokens. Meta provides BF16 and NVFP4 checkpoints for self-hosting, alongside access through hosted inference partners.The model is designed for coding, tool use and multi-step reasoning over long tasks. Its visual inputs support workflows involving screenshots, documents and other image content. Vision Evals measures its text understanding, counting, identification, reasoning and object detection at Low and High effort. The published weights use the custom Muse Model License, with commercial use subject to its terms and the accompanying usage policy.

Gemini 4 Argon is Google DeepMind's frontier Gemini 4 model, built to sustain extended reasoning across long-horizon workflows in software engineering, enterprise knowledge work such as legal and finance, scientific analysis, and defensive cybersecurity. It accepts text, image, video, and document input and returns text, operating over a one million token context window. The model raises the maximum output length to one million tokens, compared with a 64K ceiling on earlier Gemini releases, so a single reasoning trajectory can run for hundreds of thousands of tokens before terminating.On the visual side, Argon targets multimodal document comprehension, professional chart analysis, and retrieval of specific details from long recordings, reporting 91.7% on LVBench for long-video understanding and 71.6% on Chartography for chart reasoning. Google also reports 77.9% on DeepSWE v1.1, 51.3% on AutomationBench, 68% on CWE-bench v1, and first place on the Vals Index, which aggregates finance, coding, legal, and tax tasks. Following Gemini API conventions, it localizes objects by emitting bounding boxes as [ymin, xmin, ymax, xmax] integers normalized to a 0 to 1000 range, which supports prompt driven detection and grounding alongside captioning, document parsing, and visual question answering. Google describes adversarial training and automated red teaming for indirect prompt injection resistance, plus chain-of-thought and action monitors that halt execution when behavior drifts outside user intent.

Muse Wonder is a proprietary multimodal reasoning model from Meta, released on October 12, 2026 as the first step past the Muse Spark line on Meta's scaling ladder. It launched together with Muse Wonder Full, a larger configuration that is available only through Muse Code subscription plans, while Muse Wonder itself is served through the Meta Model API under the identifier muse-wonder-1.0. The model accepts text, images, video, audio, and PDF documents as input and returns text, with a context window of 1,048,576 tokens and graded reasoning effort settings. Meta trained both models at its Prometheus data center on larger base models than Muse Spark 1.3, and attributes their gains on unfamiliar tasks to post-training that transfers more effectively from a stronger base.Meta positions Muse Wonder for long-running personal and professional agents: it plans multi-step work, delegates to sub-agents, calls tools in parallel, recovers from errors, and operates browsers and desktop or mobile applications. Two training objectives shape this behaviour. Active Memory carries goals, constraints, and earlier decisions across conversations and retrieves the details relevant to the current step, and Active Focus keeps the model trying alternative approaches when a task gets difficult while asking for clarification before important actions. Visual inputs such as screenshots, charts, and video feed the same reasoning loop, and the model emits bounding box coordinates for grounding queries. Meta's model card lists visual and spatial reasoning, especially 3D tasks, as a known limitation and recommends validating performance for applications that depend on precise spatial understanding.

Claude Haiku 5.5 is a proprietary multimodal language model from Anthropic and the smallest member of the Claude 5.5 family, released on October 7, 2026 after Claude Opus 5.5 and Claude Sonnet 5.5. It accepts text and image input and returns text, with a 1M token context window and up to 128K output tokens per request. It is the first Haiku-class model with an adjustable effort parameter: adaptive thinking is on by default and the model decides how much to reason, steered by effort levels from low to max with medium as the default. Its training data cutoff is June 2026. Pricing starts at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100K tokens, which Anthropic reports is about 90% lower than Claude Haiku 4.5 for requests in that range.Anthropic positions Haiku 5.5 for high-volume, latency-sensitive work such as classification, extraction, routing, summarization, and subagent tasks, and describes it as its fastest model to date at standard speed. On visual and agentic evaluations reported at launch, it scores 46.4% on Chartography, a chart reading benchmark, compared with 6.4% for Haiku 4.5 and 61.6% for Sonnet 5.5, and 72.4% on the offline subset of OSWorld 2.1, a screenshot driven computer use benchmark, compared with 15.7% for Haiku 4.5. It uses the same tokenizer as Claude Opus 4.7 and later models, so the same text counts as roughly 30% more tokens than on Haiku 4.5.

Mistral Large 4, nicknamed Le Chonk, is a natively multimodal mixture-of-experts model from Mistral that accepts interleaved text and image input and produces text output. It uses a granular MoE design with roughly 1.05 trillion total parameters and 49 billion active per token, reported as 52 billion when embeddings and output layers are counted, paired with a 1.6 billion parameter vision encoder and a context window of one million tokens. The model is trained from scratch on about 3,800 NVIDIA Grace Blackwell GPUs in Mistral's European data centers and supports more than 160 languages. It behaves as a hybrid instruct and reasoning system, with a reasoning effort setting that selects between direct answers and longer deliberation, alongside function calling and structured output for agentic workflows.Image understanding is a focus of this generation, covering documents, charts, technical drawings and natural scenes, and the model emits bounding box coordinates for visual grounding queries. Reported grounding results include 42 percent on Dense200 and 73 percent on the DIOR-RSVG remote sensing benchmark. Mistral describes agentic vision workflows in which the model zooms into gigapixel satellite imagery or engineering drawings to verify details, and reports coding results such as 62 percent on DeepSWE. Figures published at preview time are preliminary because the reinforcement learning phase is still in progress.