Skip to content

Model detail

GPT-5.6 Luna

Provided by OpenAI
Pay-as-you-go Dynamic pricing 2 tiers

GPT-5.6 Luna is a fast, cost-efficient model in OpenAI's GPT-5.6 series. It is suited for high-volume, latency-sensitive tasks such as chat, classification, and lightweight agentic workflows, providing capable reasoning for its price tier.

Model specs

Context length
1.05M
Max output
128K
I/O modalities
Text / Image
Released
2026-07

02

API endpoints

One MixRoute gateway, OpenAI-compatible

  • OpenAI-compatible /v1/chat/completions POST

03

Tiered pricing

Unit: /1M Tokens

Tier Input /1M Tokens Output /1M Tokens Cache read /1M Tokens Cache write /1M Tokens
standard Length ≤ 272K $0.2000 $1.2000 $0.0200 $0.2500
long_context Length > 272K $0.4000 $1.8000 $0.0400 $0.5000

05

Selection summary

Quickly judge whether it fits your workload

What is this model good for?

High-volume text tasks with tools and structured output

Use GPT-5.6 Luna for high-volume chat, classification, and lightweight agent workflows that need tool calling or structured output. Start with a representative traffic slice to confirm quality and latency for your workload.

EX Run a representative chat, classification, and JSON-extraction batch with model: "gpt-5.6-luna".

What should you check before using it?

Keep requests below the 272K pricing boundary where practical

Inputs above 272K move to the higher tier: input rises from $0.20 to $0.40 and output from $1.20 to $1.80 per 1M tokens. Measure request lengths and split work where practical before production.

EX Compare a 100K-token request with a 300K-token request in your cost estimate.

Why use it through MixRoute?

Keep your OpenAI-style integration and change the route

MixRoute exposes Luna through the OpenAI-compatible chat completions endpoint, so existing OpenAI-style clients can usually keep their request format. Change the base URL and model ID, then validate tools and streaming.

EX Point an existing OpenAI-style client at MixRoute and set model: "gpt-5.6-luna".

06

Rate limits & capacity

Baseline quotas; the Console is authoritative

Available:

Requests per minute (RPM)

5,000 / min

Per-API-key request-rate ceiling for this model; exceeding it returns 429

Available:

Tokens per minute (TPM)

4M tokens / min

Combined input + output tokens; varies by plan tier and can be raised on request

Available:

Automatic failover on 429

On provider rate limiting, traffic automatically shifts to another available channel; no client-side retry logic needed

EX When the primary channel returns 429 at peak, the system instantly resends over a backup channel; responses are unaffected and billed at the same rate

07

Capabilities & compatibility

Unconfirmed capabilities stay "Unknown," no guesses

Available:

Tool calling

Supports function and tool calls

EX Send a tools definition; the response returns tool_calls to execute and feed results back
Available:

Structured output

Can require JSON-formatted output

EX response_format: { type: "json_schema" } forces valid JSON output
Available:

Streaming

Supports streamed responses

EX stream: true streams tokens incrementally, so chat UIs render as they arrive
Conditional:

Prompt caching

Depends on the provider and routing conditions

EX Keep the system prompt and tool definitions at the start of the request; repeated prefixes are automatically billed at the cache-read rate
Unknown:

Vision input

Official capability data is unconfirmed; listed as unknown for now

Available:

OpenAI SDK

Call it through MixRoute's compatible endpoint

EX Change base_url to https://api.mixroute.ai/v1; everything else follows the OpenAI style

08

Token cost estimator

Live estimate from this page's pricing, not an actual bill

Share of the same prefix read repeatedly, up to 100%

09

FAQ

Cache read is $0.0200 / 1M tokens, only one tenth of the $0.2000 input price, but cache write costs $0.2500, 25% more than input itself. That means a prompt written once and then discarded actually costs more. The test is how many times the same prefix gets read again — content sent on every request, like system prompts, tool definitions, and fixed knowledge, is worth caching; one-off long documents are not. Before rolling it out, measure what share of your requests are repeated prefixes, then decide whether to enable it.

One endpoint, a testable decision

Test this model and alternate routes with the same request format

Start from a real workload, then let quality, total cost, and failure conditions decide whether to send production traffic.