Access OpenAI models and tools through SandBase.
45 available model and capability records—not 45 independent foundation models.
Catalog verified · Oct 13, 2026 · Production provider source · Live model catalog
A Compact Starting Set
Selected from the current catalog.
Not a benchmark ranking.
openai/gpt-6.1-solGPT-6.1 Sol is OpenAI's upgrade to GPT-6 Sol, introduced at DevDay 2026 and positioned between GPT-6 Astra and GPT-6 Luna, with text and image input. OpenAI says it delivers near-Astra performance for...
1,050,000 context tokens
openai/gpt-6-solGPT-6 Sol is the middle tier of OpenAI's GPT-6 reasoning family, released in September 2026 alongside GPT-6 Luna and built, according to OpenAI, for complex coding and agentic workflows, with text and...
1,050,000 context tokens
openai/gpt-6-lunaGPT-6 Luna is the smallest, fastest tier of OpenAI's GPT-6 reasoning family, described by OpenAI as its most efficient model for focused, high-volume tasks, with text and image input. It was trained w...
1,050,000 context tokens
openai/gpt-6-astraGPT-6 Astra is OpenAI's flagship GPT-6 reasoning model, described by OpenAI as its most capable model for the most demanding work, with text and image input. OpenAI recommends it for complex reasoning...
1,050,000 context tokens
openai/gpt-5.3-codexGPT-5.3-Codex is OpenAI's agentic coding model announced in February 2026, optimized for Codex and similar environments, with text and image input and text output. OpenAI called it its most capable ag...
400,000 context tokens
openai/gpt-image-2.5-flareGPT Image 2.5 Flare is optimized for fast, high-quality image generation for everyday creative workflows.
Final cost can vary by duration, resolution, quality, or inputs.
LLM prices are per million input/output tokens; embeddings use input tokens. Request prices are base prices. Long-context, caching, duration, or other request inputs may change final cost. Confirm the live model page and schema.
OpenAI / llm
Chat, reasoning, structured output, and function calling. Inspect each record for its supported capabilities and context window.
openai/gpt-6.1-solGPT-6.1 Sol is OpenAI's upgrade to GPT-6 Sol, introduced at DevDay 2026 and positioned between GPT-6 Astra and GPT-6 Luna, with text and image input. OpenAI says it delivers near-Astra performance for...
1,050,000 context tokens
openai/gpt-6-solGPT-6 Sol is the middle tier of OpenAI's GPT-6 reasoning family, released in September 2026 alongside GPT-6 Luna and built, according to OpenAI, for complex coding and agentic workflows, with text and...
1,050,000 context tokens
openai/gpt-6-lunaGPT-6 Luna is the smallest, fastest tier of OpenAI's GPT-6 reasoning family, described by OpenAI as its most efficient model for focused, high-volume tasks, with text and image input. It was trained w...
1,050,000 context tokens
openai/gpt-6-astraGPT-6 Astra is OpenAI's flagship GPT-6 reasoning model, described by OpenAI as its most capable model for the most demanding work, with text and image input. OpenAI recommends it for complex reasoning...
1,050,000 context tokens
openai/gpt-5.6-solGPT-5.6 Sol is the flagship tier of OpenAI's GPT-5.6 reasoning family, accepting text and image input; OpenAI says it roughly corresponds to the unsuffixed tier of earlier GPT-5 families, and the plai...
1,050,000 context tokens
openai/gpt-5.6-terraGPT-5.6 Terra is the middle tier of OpenAI's GPT-5.6 reasoning family, with text and image input, and is described by OpenAI as a balanced model for everyday work; it roughly corresponds to the mini t...
1,050,000 context tokens
openai/gpt-5.6-lunaGPT-5.6 Luna is the smallest, fastest tier of OpenAI's GPT-5.6 reasoning family, with text and image input; OpenAI says it roughly corresponds to the nano tier of earlier GPT-5 families. It is designe...
1,050,000 context tokens
openai/gpt-5.5GPT-5.5 is an OpenAI flagship reasoning model with text and image input, made available in the API in April 2026 and described by OpenAI as built for complex professional work. OpenAI positions it for...
1,050,000 context tokens
openai/gpt-5.4GPT-5.4 is OpenAI's flagship GPT-5.4 reasoning model for complex professional work, released in March 2026 with text and image input, text output and a context window of about 1.05M tokens. OpenAI pos...
1,050,000 context tokens
OpenAI / image
Separate text-to-image and image-to-image records preserve the model and task-specific action.
openai/gpt-image-2.5-flareGPT Image 2.5 Flare is optimized for fast, high-quality image generation for everyday creative workflows.
Final cost can vary by duration, resolution, quality, or inputs.
openai/gpt-image-2.5-flare/editGPT Image 2.5 Flare Edit is optimized for fast, high-quality image editing and multi-image composition.
Final cost can vary by duration, resolution, quality, or inputs.
openai/gpt-image-2.5-sunburstGPT Image 2.5 Sunburst prioritizes visual fidelity and precision for complex image generation workflows.
Final cost can vary by duration, resolution, quality, or inputs.
openai/gpt-image-2.5-sunburst/editGPT Image 2.5 Sunburst Edit prioritizes precision and visual fidelity for complex image editing workflows.
Final cost can vary by duration, resolution, quality, or inputs.
openai/gpt-image-2GPT Image 2, OpenAI's latest image model, is capable of making fine-grained, detailed edits to images.
Final cost can vary by duration, resolution, quality, or inputs.
openai/gpt-image-2/editGPT Image 2 Editing supports image editing and multi-image synthesis with high-quality results.
Final cost can vary by duration, resolution, quality, or inputs.
OpenAI / audio
The current OpenAI audio record is Wizper for speech-to-text transcription; this inventory does not establish speech-generation availability.
openai/wizperWizper by OpenAI - accurate speech-to-text transcription with AI. Convert audio and video to text with high accuracy, multilingual support, and speaker identification.
Final cost can vary by duration, resolution, quality, or inputs.
OpenAI / embedding
Create text embeddings for semantic search, retrieval, clustering, and classification.
openai/text-embedding-3-largeOpenAI text-embedding-3-large creates embeddings for semantic search, retrieval, clustering, and classification.
8,191 context tokens
Models Are Runtime Configuration
The Agent owns the task; the selected OpenAI model is swappable, not permanently bound to that Agent. Skills, MCP/Tools, and real-world APIs remain separate configuration. Check tool support and schema compatibility before switching.
Choose by Workload
The catalog describes Astra as the flagship GPT-6 tier for demanding work; this is positioning, not a measured performance ranking.
The catalog describes Terra as a balanced model for everyday work. Validate it against your task and latency needs.
The catalog positions Luna for focused, high-volume tasks. Compare its input/output rates with your expected usage.
The catalog describes GPT-5.3-Codex as an agentic coding model; verify tool and schema compatibility.
Flare has separate generation and editing records. Compare Sunburst for the catalog's visual-fidelity positioning.
Wizper is tagged speech-to-text. Confirm duration-based billing and accepted media on its live model page.
The catalog lists semantic search and retrieval. Pricing uses input tokens, not generated-output tokens.
Catalog positioning is a starting point, not an independent performance guarantee. Compare the record cards for units and context, then evaluate your own workload.
Keep Exploring
50 models & variantsLanguage, visual creation, video, speech, and music models.
16 models & variantsClaude language models for reasoning, writing, and coding.
8 models & variantsLanguage models for reasoning and general text tasks.
20 models & variantsGrok language models and Grok Imagine visual generation.
FAQ
Choose an OpenAI model record, then open its SandBase model page or Playground for the current request schema and action. SandBase provides one integration and shared API key; the production OpenAI provider page remains a reference source.
No. The Oct 13, 2026 catalog has 45 enabled model and capability records: 34 LLM, 9 image, 1 audio, and 1 embedding. Task variants and pro-mode records are included, so the total is not a count of independent foundation models.
Many LLM records carry a vision tag. Image records distinguish generation from editing, Wizper provides speech-to-text, and text-embedding-3-large provides text embeddings. These records do not establish video generation or text-to-speech availability.
LLM prices show input and output rates per million tokens. Embeddings use input-token pricing. Image and audio cards show base request prices; formulas, inputs, cached tokens, long-context thresholds, or duration can affect final cost. Always confirm the live model page before use.
Use each record's catalog context value rather than assuming every OpenAI model has the same limit. For example, GPT-6.1 Sol lists 1,050,000 tokens, GPT-5.3-Codex lists 400,000, and GPT-4o mini lists 128,000. A context value is not a guaranteed maximum output length.
Start with your workload, supported tags, context requirement, and pricing unit. The comparison section uses catalog positioning as a starting point, not benchmark rankings. Test your own prompts, tools, and cost constraints before selecting a production model.
Yes, treat the selected model as swappable runtime configuration rather than a permanent Agent dependency. Recheck input/output behavior, context, tool support, and request schema whenever switching models.
An Agent owns the task. Its selected OpenAI model works with separately configured Skills and MCP/Tools, which can call real-world APIs. A model record does not by itself prove access to any particular Skill, tool, or API.