LLMAnonymized

GPT-5.6 Luna Pro

OpenAI's cost-efficient GPT-5.6 tier with a 1M context, vision, reasoning, and tool use for high-volume workloads.

Maker
OpenAI
Modality
Text + vision
Context
1,000K tokens
Open weights
No

Overview

What is GPT-5.6 Luna Pro

GPT-5.6 Luna Pro is OpenAI's efficient, high-volume tier in the GPT-5.6 family, released in July 2026. It offers a 1M-token context, vision, reasoning, and tool use at a lower per-token cost than Sol and Terra, making it ideal for production workloads that need frontier capability without flagship pricing.

Running it privately on Venice

On Venice, GPT-5.6 Luna Pro runs under an anonymized privacy tier — your prompts are not stored, profiled, or used for training. You get the same OpenAI frontier model with vision, reasoning, and tool use, but without the surveillance footprint of Big-Tech platforms. It is available via the Venice app and API with zero retention.

AnonymizedNo prompt trainingTEE · hardware enclaveEnd-to-end encrypted

Assessment

Strengths and limitations

Strengths
  • 1M-token context window and 128K max output, enabling long-document analysis and large codebases.
  • Native vision, reasoning, tool use, and web search for multimodal agentic workflows.
  • Most cost-efficient GPT-5.6 tier, optimized for high-throughput production workloads.
  • Supports structured JSON output and multiple image inputs for reliable integration.
Limitations
  • Closed weights: cannot be self-hosted, fine-tuned, or audited.
  • Retains OpenAI content policies; not uncensored, so refusals may occur on sensitive topics.
  • Output token cost is steep compared to open-weights rivals on Venice.
  • Privacy tier is anonymized, not end-to-end encrypted or TEE-hardened.

Capabilities

What it supports

  • Tool use / function calling
  • Vision (image input)
  • Reasoning
  • Web search
  • Code-optimized
  • Structured output (JSON schema)
  • Audio input
  • Video input
  • Multiple image inputs
  • Log probabilities

Specifications

Datasheet

Maker
OpenAI
Released
July 9, 2026
License
Proprietary
Open weights
No
Context window
1,000K tokens
Max output
128K tokens
Capabilities
Vision, Function calling, Reasoning, Web search
Privacy on Venice
Anonymized — prompts not stored
Available on Venice since
Jul 2026

API

Call it from your code

Venice exposes an OpenAI-compatible API. Point your base URL at Venice and pass the model id.

curl https://api.venice.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai-gpt-56-luna-pro",
    "messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
  }'

Pricing

What it costs on Venice

Billed per token on Venice: $1.25 per 1M input tokens and $7.50 per 1M output tokens.

Input / 1M tokens
$1.25
Per 1M tokens
Output / 1M tokens
$7.50
Per 1M tokens
Cached input / 1M
$0.13
Per 1M tokens

New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.

Alternatives

How it compares

ModelContext windowPrice (Venice)Open weightsBest for
GPT-5.6 Luna Pro1M tokens$1.25 in · $7.50 out / 1MNoEfficient high-volume workloads
Claude Sonnet 4.61M tokens$3.60 in · $18 out / 1MNoBalanced performance
DeepSeek V3.2160K tokens$0.33 in · $0.48 out / 1MYesOpen cheap inference
Grok 4.31M tokens$1.42 in · $2.83 out / 1MNoReal-time queries

The most cost-efficient GPT-5.6 tier with 1M context, vision, and tool use — run privately with zero prompt retention.

Use cases

What it is good for

  1. 01High-volume API workloads like summarization, classification, and data extraction.
  2. 02Long-context document Q&A and legal or medical record analysis.
  3. 03Vision-enabled agents that parse screenshots, charts, and PDFs.
  4. 04Coding assistance and multi-file repository reasoning.
  5. 05Structured data generation with guaranteed JSON schema compliance.

Prompting

Getting better results

Mark reusable prompt prefixes for caching to reduce costs on repetitive workflows.

Start with medium reasoning and test one level lower; Luna Pro often maintains quality with fewer tokens.

Use explicit JSON schema constraints when you need structured output.

For vision tasks, upload high-resolution images directly and reference specific regions in your prompt.

Version history

GPT-5.6 Sol
2026-07

Flagship tier with maximum reasoning capability.

GPT-5.6 Terra
2026-07

Balanced tier for everyday work.

GPT-5.6 Luna Pro
2026-07

Current efficient tier optimized for high-volume workloads.

FAQ

Frequently asked questions

GPT-5.6 Luna Pro is OpenAI's efficient, high-volume tier in the GPT-5.6 family, released in July 2026. It delivers a 1M-token context, vision, reasoning, and tool use at a lower per-token cost than the Sol and Terra tiers.

On Venice, GPT-5.6 Luna Pro costs $1.25 per 1M input tokens and $7.50 per 1M output tokens. Cached input is billed at $0.13 per 1M tokens. You pay only for what you use with no subscription required.

No. GPT-5.6 Luna Pro is proprietary closed-source software from OpenAI. Its weights are not available for download, self-hosting, or fine-tuning.

No, it consumes credits on Venice. New accounts receive welcome credits that let you try it before purchasing more.

Luna Pro offers a 1M context and vision at $1.25/$7.50 per 1M tokens, while Claude Sonnet 4.6 costs $3.60/$18 per 1M. Both are closed weights. Choose Luna Pro for cost-efficient long-context and multimodal work; Sonnet 4.6 if you prefer Anthropic's policy tuning.

Yes. On Venice it supports function calling, structured JSON output, web search, and multiple image inputs for building agentic workflows.

Yes. It accepts vision inputs including multiple images, and can reason over charts, screenshots, and documents alongside text.

It can generate up to 128K tokens in a single response, making it suitable for long-form writing and large code generation.

No. Venice runs the model with its original OpenAI weights and safety filters intact. It is not flagged as uncensored, so standard content policies still apply.

Run GPT-5.6 Luna Pro privately

No prompt logging. No data used for training.