How much do GPT-6 Sol and GPT-6 Luna cost?

GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, with cached input at $0.20 and cache writes at $2.50. GPT-6 Luna costs $0.10 input and $0.50 output, with cached input at $0.01. Both are 50% cheaper than the promotional prices of GPT-5.6 Sol ($4/$20) and GPT-5.6 Luna ($0.20/$1.20). Prompts over 272,000 input tokens are billed at 2x input and 1.5x output for the whole request; Batch and Flex are 50% off; fast mode is 2x.

Is GPT-6 Sol better than GPT-5.6 Sol?

OpenAI says yes on professional work, factuality, coding and computer use: GPT-6 Sol makes about half as many factual mistakes as GPT-5.6 Sol on OpenAI's internal evaluation, scores 33.2% on AutomationBench at xhigh (beating Claude Opus 5 at 9% of its cost per task), 56.4% on Agents' Last Exam and 68.8% on DeepSWE v1.1 at max, within 1.1 points of Claude Fable 5's best. Artificial Analysis is more cautious: it measured Sol's Intelligence Index rising only from 47 to 48, with some evaluations improving and others regressing, and calls the release mainly a cost story.

What happened to GPT-6 Terra?

There is no GPT-6 Terra as of September 23, 2026. OpenAI's GPT-6 lineup is Astra ($10/$50), Sol ($2/$10) and Luna ($0.10/$0.50). GPT-5.6 Terra remains available at $2/$12, which is now more expensive than GPT-6 Sol at the same input price, so there is little reason to keep using it. GPT-5.6 Sol, Terra and Luna are also due for a scheduled price increase in November 2026 when their promotional pricing ends.

Quick Answer

What Are GPT-6 Sol and GPT-6 Luna? OpenAI's Sep 2026 Cuts

Q: What are GPT-6 Sol and GPT-6 Luna?

GPT-6 Sol and GPT-6 Luna are the two smaller members of OpenAI's GPT-6 family, released September 22, 2026, nineteen days after GPT-6 Astra. Sol is built for complex coding and agentic workflows; Luna is OpenAI's most efficient model for focused, high-volume tasks. Both were trained with similar methods to Astra, share its 1,050,000-token context window and 128,000 max output, take text and image input, support reasoning effort from none to max, and are available as gpt-6-sol and gpt-6-luna in the API and in ChatGPT Work and Codex for paid plans.

Published: September 23, 2026

The short answer

GPT-6 Sol and GPT-6 Luna are OpenAI’s cheaper GPT-6 models, launched September 22, 2026 at exactly half the price of the GPT-5.6 models they replace: Sol at $2/$10 per million tokens and Luna at $0.10/$0.50. They inherit GPT-6 Astra’s training recipe, context window and communication style, target “work at different scales, rhythms and budgets,” and arrived about an hour after Anthropic’s Claude Opus 5.5. The price cut is the headline; the intelligence gain over GPT-5.6 is real but modest by independent measurement.

At a glance

	GPT-6 Sol	GPT-6 Luna
Model ID	`gpt-6-sol`	`gpt-6-luna`
Positioning	Complex coding and agentic workflows	Focused, high-volume tasks
Price (in / cached / out per MTok)	$2 / $0.20 / $10	$0.10 / $0.01 / $0.50
Cache write	$2.50	$0.125
Previous model	GPT-5.6 Sol at $4 / $0.40 / $20	GPT-5.6 Luna at $0.20 / $0.02 / $1.20
Context / max input / max output	1,050,000 / 922,000 / 128,000	Same
Knowledge cutoff	April 20, 2026	May 18, 2026
Modalities	Text + image in, text out	Same
Reasoning effort	none, low, medium (default), high, xhigh, max	Same
>272K-token prompts	2x input and cache rates, 1.5x output	Same
Batch / Flex / fast mode	50% / 50% / 2x	Same
Endpoints	Responses, Chat Completions, Batch (no Realtime, Live, fine-tuning)	Same
Tools (Responses API)	web_search, file_search, image_generation, code_interpreter, hosted_shell, apply_patch, skills, computer_use, mcp, tool_search	Same
Tier-5 rate limit	15,000 RPM / 40M TPM	30,000 RPM / 180M TPM

Where they are available

API: gpt-6-sol and gpt-6-luna, Responses and Chat Completions (function calling in Chat Completions only with reasoning_effort: none). EU data residency with Standard processing only.
ChatGPT Work and Codex: all Plus, Pro, Business, Enterprise and Edu users, rolling out through September 22.
Free and Go: GPT-6 Luna in the desktop app.
Not yet in regular ChatGPT Chat as of launch day.

Benchmarks OpenAI published

Evaluation	GPT-6 Sol	Comparison
AutomationBench 1.0.6 (Zapier)	33.2% at xhigh, $0.27/task	Beats Claude Opus 5 max (26.9%) at 9% of its cost; beats GPT-6 Astra low (30.3%) at ~1/4 the cost; Fable 5.1 w/ fallback 31.4%
Agents’ Last Exam V1	56.4% at max	Above Claude Opus 5’s best at 60% lower cost per task
DeepSWE v1.1	68.8% at max	Within 1.1 pts of Fable 5’s 69.9% at ~80% lower cost
FrontierCode 1.1 Main	Matches Fable 5.1 xhigh	At “much lower cost” (no figure given)
OSWorld 2.0 offline	60.5% at xhigh	vs Opus 5 medium 60.3% at ~80% lower cost
Internal factuality	~50% fewer mistakes than GPT-5.6 Sol	”Approaching Astra-level reliability”

Evaluation	GPT-6 Luna	Comparison
AutomationBench	+5.4 pts over GPT-5.6 Luna at high	58% lower cost per task
DeepSWE v1.1	66.6% at max	Comparable to Opus 5 / Fable 5 at medium; 93–96% cheaper
OSWorld 2.0 offline	Luna max > GPT-5.6 Sol medium	At one tenth the cost
Factuality	Matches GPT-5.6 Sol at higher effort	At ~1/100th the cost

Note that OpenAI’s comparisons are against Claude Opus 5, not the Claude Opus 5.5 that launched the same day, and that “scores for Claude Fable 5 were reported when scores for Claude Fable 5.1 were unavailable.”

The independent view

Artificial Analysis summarised the release as OpenAI “halving cost relative to GPT-5.6 Sol and Luna” while “Intelligence Index and Coding Agent Index scores remain level with GPT-5.6, with progress in some evaluations and regressions in others.” Sol moved from 47 to 48 on the current index scale; Claude Opus 5.5 scored 58 the same day. So GPT-6 Sol is not a frontier model; it is a cheaper GPT-5.6 Sol with Astra’s writing style and better caching.

That framing matters for the Opus 5.5 vs GPT-6 Sol decision: Sol is half Opus 5.5’s list price but ten index points behind.

Better caching for the whole GPT-6 family

Alongside the models, OpenAI shipped an improved prompt-caching system: higher hit rates by default, cache discounts for shared prefixes reused within a 30-minute window, a Prompt Caching Dashboard, a diagnostics tool that explains misses ("reason": "tools_changed", with the token count affected), explicit cache breakpoints, cache prewarming, and the ability to change reasoning effort or toggle tools mid-conversation without breaking the cache. GitHub says these changes cut the share of prompt tokens needing fresh processing by more than 50% across Copilot’s requests. See How to maximize prompt cache hit rate.

What it means for the GPT-5.6 lineup

GPT-5.6 Terra ($2/$12) now costs more than GPT-6 Sol for the same input price and a weaker model. As Simon Willison put it, “any remaining reasons to use Terra just evaporated.”
GPT-5.6 Sol ($4/$20) is now the same price as Claude Opus 5.5, a stronger model, and twice the price of GPT-6 Sol.
GPT-5.6 Luna ($0.20/$1.20) is twice GPT-6 Luna’s price. Luna was already the free ChatGPT default; the GPT-6 version now reaches Free and Go users on desktop.
The GPT-5.6 promotional prices are scheduled to rise in November 2026, which widens every gap above.

The GPT-5.5 retirement on October 14, 2026 is a separate event; GPT-5.6 models remain available.

Who should switch

Building on GPT-5.6 Terra or Sol: switch to GPT-6 Sol now; same or lower price, better factuality, Astra-style output.
High-volume classification, extraction, chat: GPT-6 Luna at $0.10/$0.50 undercuts DeepSeek V4.1 Flash, Gemini 3.8 Flash and Claude Haiku 4.5 ($1/$5) by a wide margin. Test quality on your data; AA’s index says it is roughly a GPT-5.6-class model, not a leap.
Need frontier reasoning: stay on GPT-6 Astra or move to Claude Opus 5.5. See GPT-6 Sol vs Luna vs Astra.

Last verified: September 23, 2026.