AI library

Models

GPT (OpenAI)

Last reviewed 16 October 2026

Overview

OpenAI's GPT models power ChatGPT, Codex, and the OpenAI API. GPT-5 (August 2025) merged the separate GPT and "o-series" reasoning lines into one family. A series of GPT-5.x releases followed through 2026, and GPT-6 arrived in September 2026 in three tiers: Astra in early September, then Sol and Luna at API prices 50% below GPT-5.6 promotional pricing. GPT-6.1 Sol replaced GPT-6 Sol on 2026-09-29.

Key concepts

  • GPT-6 tiers: Astra (gpt-6-astra, flagship; OpenAI claims state of the art on computer use, browsing, software engineering, cybersecurity, and professional work), Sol (gpt-6.1-sol, near-Astra quality at lower cost), Luna (gpt-6-luna, cheapest and fastest).
  • Reasoning built in: reasoning effort is configurable instead of being a separate model line.
  • Surfaces: ChatGPT, Codex (coding agent), and the API (Responses API, Agents API, Agents SDK, and the Decisions API for typed classification answers).
  • Usage tiers (since 2026-10-06): Free ($100 a month in eligible countries), then Build (after $5 of credit purchases, $500 a month), Launch (after $100, $5,000 a month), and Grow (after $500, $200,000 a month). These replaced the earlier five tiers. Organisations move up automatically as their total credit purchases grow.
  • Hosted agents: the Responses API supports multi-agent runs (beta), and the Agents API has hosted computer use (both since 2026-09-29).

How to use it

  • Pick a tier by cost and difficulty, as with other vendors.
  • All GPT-6 models have a 1.05M-token context window and 128K output (as of 2026-10-06), but inputs above 272K tokens cost 2x for input and 1.5x for output.
  • Prices per 1M tokens (input / cached input / output, as of 2026-10-06):
Model Price Knowledge cutoff
GPT-6 Astra $10 / $1.00 / $50 2026-04-30
GPT-6.1 Sol $2 / $0.10 / $10 2026-04-30
GPT-6 Luna $0.10 / $0.01 / $0.50 2026-05-18
  • For classification, routing, or yes/no checks, the Decisions API (POST /v1/decisions, gpt-6-luna only, beta since 2026-10-06) answers predicate, choice, or score questions about text and inline images. OpenAI says it is about 10x faster than the Responses API. It charges $0.10 per 1M input tokens and nothing for output (as of 2026-10-07).
  • For lower latency between output tokens, set service_tier: "ultrafast" in the Responses API on gpt-6-astra or gpt-6.1-sol (GPT-5.6 Sol in preview). It costs more and has its own rate limits; OpenAI recommends WebSockets for agents. Astra's Ultrafast has no EU data residency (as of 2026-10-09).

Current state

  • 2026-10-15: GPT-Live (gpt-live-1) now integrates natively with the Decisions API: it can delegate to Decisions directly, or Decisions calls can fire on transcript deltas, so a voice app can classify and act while the user is still speaking (OpenAI API changelog, entry of 2026-10-15; no pricing or latency figures given).
  • 2026-10-08: Ultrafast mode added for GPT-6.1 Sol in the Responses API, for all API users, with global processing and US or EU data residency.
  • 2026-10-06: Decisions API released in beta (details above).
  • 2026-10-06: API usage tiers cut from five to three (Build, Launch, Grow).
  • 2026-10-05: Wikimedia reported unauthorised edits, exploit attempts, and heavy traffic from agents it attributes to OpenAI (see security).
  • 2026-09-29: GPT-6.1 Sol released (confirmed by OpenAI's API changelog), with cheaper cached input than GPT-6 Sol ($0.10 vs $0.20). GPT-6 Astra gained the ultrafast service tier the same day.
  • 2026-09-25: OpenAI fixed an image-encoding bug that had degraded image understanding in GPT-6 Sol and Luna; rerun image evals made before the fix.

Open questions

  • The exact release date of GPT-6 Astra; the Sol and Luna announcement says only "early September 2026".
  • When was the Decisions API first announced? Secondary coverage says DevDay (2026-09-29); OpenAI's changelog gives the beta release date as 2026-10-06. Latency figures such as ~150 ms come only from secondary reports.

Sources