OpenAI’s July 2026 model generation with three durable tiers: Sol (flagship), Terra (balanced everyday work), and Luna (fastest and cheapest). The generation number and tier names can evolve separately.

When did GPT-5.6 launch?

General availability across ChatGPT, Codex, and the API began 9 July 2026 after a limited preview. Rollout was gradual. On 30 July 2026 OpenAI cut API prices for Terra and Luna and documented Fast mode for Sol.

How much does GPT-5.6 cost on the API?

As of OpenAI’s 30 July 2026 post and API pricing docs: Sol remains $5 input / $30 output per 1M tokens (standard short-context); Terra $2 / $12; Luna $0.20 / $1.20. Cached reads keep a large discount; cache writes are billed higher than uncached input on 5.6-class models. Always re-check developers.openai.com pricing.

Which ChatGPT plans get Sol?

OpenAI states Plus, Pro, Business, and Enterprise users access GPT-5.6 Sol via medium and higher effort settings; Pro and Enterprise can also select Sol Pro for highest-quality complex work. Terra/Luna availability differs by product (Work, Codex, API) — Free/Go Codex gets Terra per OpenAI’s price-cut post.

Is Terra a replacement for GPT-5.5?

OpenAI positions Terra as competitive with GPT-5.5-class performance at lower cost after the July cut. For production, run your own eval suite before swapping defaults — list prices and marketing parity are not the same as your task mix.

Does this change Codex billing?

Subscription list prices and quota budgets for ChatGPT/Codex were described as unchanged on 30 July, while Terra and Luna usage consume fewer credits after the cut. Sol Fast mode aligns with faster API processing at 2× standard price.

GPT-5.6 Sol, Terra, Luna: What Changed and Pricing

GPT-5.6 is OpenAI’s mid-2026 model generation: not one model, but three named capability tiers — Sol, Terra, and Luna — shipped for ChatGPT, Codex, and the API starting 9 July 2026, with a material Terra/Luna price cut on 30 July 2026.

This is a PromptHive launch analysis. Facts below are taken from OpenAI’s public launch and pricing posts and API pricing tables. We translate them into decisions for people already paying for ChatGPT, building on the API, or choosing coding agents.

Verified 2026-08-05 against OpenAI’s GPT-5.6 launch page, the 30 July price-performance post, and public API pricing tables. Re-check official pricing before you budget.

The short answer

Sol — flagship reasoning/agent work; API ~$5 / $30 per 1M tokens (short-context standard); highest-quality ChatGPT effort paths on paid plans.
Terra — everyday / balanced tier; after 30 July: API $2 / $12; OpenAI’s “GPT-5.5 competitive at lower cost” story.
Luna — high-volume / speed tier; after 30 July: API $0.20 / $1.20; OpenAI’s cheapest 5.6-class model.
Subscription seats — ChatGPT/Codex plan prices were not rewritten as a new SKU on 30 July; Terra/Luna burn fewer credits after the cut.
PromptHive take: use Sol when the task is hard and review is expensive; default Terra for bulk knowledge work; Luna for high-QPS agents. Measure your own harness.

What was officially announced

9 July 2026 — GPT-5.6 general availability

OpenAI launched the GPT-5.6 family for general availability across ChatGPT, Codex, and the OpenAI API after a limited partner preview. Naming rule (official):

Piece	Meaning
5.6	Generation
Sol / Terra / Luna	Durable capability tiers that can advance on their own cadence

Initial list rates at launch (per OpenAI launch materials): Sol $5/$30, Terra $2.50/$15, Luna $1/$6 per 1M tokens, plus new prompt-caching rules (cache writes at 1.25× uncached input; 90% discount on cache reads for 5.6-class models).

30 July 2026 — Terra and Luna cheaper; Sol Fast mode

OpenAI’s “price-performance frontier” post cut:

Model	New API list (short-context, per OpenAI)	Change vs launch list
Sol	$5 / $30 (unchanged)	—
Terra	$2 / $12	~20% cheaper than $2.50/$15
Luna	$0.20 / $1.20	~80% cheaper than $1/$6

Also documented: Fast mode for Sol (up to ~2.5× standard speed at 2× standard price), aligning with faster processing in API/Codex. ChatGPT and Codex subscription prices and quota budgets were stated as unchanged, while Terra/Luna usage consumes fewer credits.

Public API pricing tables (developers.openai.com) list matching short-context rates for gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna, with higher long-context rows.

What’s new vs the previous generation

Dimension	Prior mental model	GPT-5.6
Product shape	Single flagship + smaller siblings	Explicit three-tier family with durable names
Cost control	Often “one expensive model”	Luna for volume; Terra for daily; Sol for hard
ChatGPT routing	Model picker / effort	Sol via medium+ effort on paid; Sol Pro on Pro/Enterprise
Codex	Bundle with plan	Terra for Free/Go; full tier set on paid (per OpenAI)
Caching	Cached-input discounts	Explicit breakpoints + 30-minute minimum cache life called out for 5.6

Features explained

Sol — when intelligence and review time dominate

Sol is OpenAI’s flagship 5.6 tier for complex reasoning, coding, science, and long-horizon agent work. On ChatGPT, paid users reach it through effort settings rather than always seeing a simple radio button for every sub-variant.

Use Sol when: a wrong answer costs more engineer time than tokens. Pair with git discipline if you use Codex.

Terra — default workhorse

Terra is the balanced tier for everyday professional work. After the 30 July cut it is the practical default for many API and Work/Codex workloads that do not need Sol.

Use Terra when: you want strong quality without Sol’s bill, and you have tests or humans catching errors.

Luna — volume and latency

Luna is the speed/cost tier for high-throughput agents and multi-step workflows where OpenAI claims strong quality at low token prices.

Use Luna when: QPS and unit economics matter more than peak single-task score.

Fast mode (Sol)

Pay 2× Sol’s standard rate for up to ~2.5× speed. Same intelligence claim; different wall-clock and bill. Do not confuse Fast mode with Luna.

Pricing and availability (decision table)

Channel	What OpenAI states (public)
API	Sol / Terra / Luna with rates above; long-context higher
ChatGPT consumer	Sol via medium+ effort on Plus/Pro/Business/Enterprise; Sol Pro option on Pro/Enterprise
ChatGPT Work	Sol, Terra, Luna on Plus/Pro/Business/Enterprise (per help materials)
Codex	Terra on Free/Go; Sol+Terra+Luna on paid tiers (per OpenAI 30 July post)
Subs list prices	Unchanged on 30 July cut; credits stretch further on Terra/Luna

Trap: “Cheaper Luna” does not mean “free infinite agents.” Caps, effort, tools, and Fast mode still dominate the bill.

PromptHive hands-on analysis

We did not re-score every public leaderboard. Practical reading for PromptHive readers:

This is a packaging win. Three named tiers reduce the old “is this mini or full?” confusion — if product UIs actually expose them clearly in your account.
The 30 July cut matters more than the 9 July sticker for Terra/Luna API users. Re-run unit economics if you priced agents in early July.
Sol vs Claude Opus 5 is the real premium coding comparison, not Sol vs Luna. See Claude Opus 5 and Codex vs Claude Code.
Codex users should treat Terra as the free/Go default and Sol as the paid heavy path — then measure remaining quota after one real ticket.
Safety and rollout were part of the story (system card, gradual availability). If a model is missing from your picker, assume rollout lag before assuming a plan bug.

Pros

Clear three-tier family with official cost ladder after 30 July
Luna makes high-volume agent APIs more realistic on paper
Sol Fast mode is an explicit latency purchase, not a hidden overage
Tight integration story across ChatGPT, Codex, and API
Caching rules documented for 5.6-class models

Cons

Plan × product × effort matrix is still complex to explain to teams
Vendor benchmarks are not your production suite
Sol still expensive if used as the default for everything
Gradual rollouts create “why don’t I see it?” support load
Competing stacks (Claude Opus/Fable, Gemini Flash, Grok 4.5) move monthly

Best use cases

Job	Start with
Hard multi-file agent ticket (paid)	Sol (standard or Fast if waiting is costlier)
Everyday drafting, analysis, light coding	Terra
Classification, routing, high-QPS tools	Luna
ChatGPT free curiosity	Free plan paths OpenAI documents — not Sol as unlimited
Claude-only shops	Stay on Anthropic; compare only if multi-homing

30-minute evaluation plan

Pick one real task you already run (agent ticket, research memo, or classification batch).
Run it on Terra with your normal tools/context.
Run the hard version on Sol (standard, not Fast, first).
If latency is the bottleneck on Sol, retry with Fast mode and compare cost.
For volume jobs, run a 50–100 item sample on Luna and score error rate × $/item.
Record: dollars (or credits), minutes of human review, and whether you would ship.

That experiment beats another week of reading leaderboard screenshots.

How this interacts with Claude and Gemini

Stack	When GPT-5.6 wins	When the other wins
Claude Opus 5 / Claude Code	You are already multi-homed on OpenAI; Codex cloud tasks; Luna volume	You standardise on Anthropic agents and `CLAUDE.md` culture — see Opus 5
Gemini Flash line	You need ChatGPT product surface + Codex	You need Google ecosystem agents / cheap Flash scale — see Gemini 3.6 Flash
Grok 4.5	You want OpenAI’s product depth	You want $2/$6 coding-oriented API and Cursor-adjacent training story — see Grok 4.5

Final PromptHive verdict

GPT-5.6 is less a single “new ChatGPT brain” and more a productized ladder: Sol for hard work, Terra for daily work, Luna for volume — with a late-July cut that actually changes API unit economics. If you already pay for OpenAI, re-default agents to Terra and keep Sol for the tickets where review time is the real bill. If you live in Claude Code, treat this as competitive pressure, not an automatic migration.

CTA: Rebuild one production prompt on Terra and one on Sol; compare cost × review minutes. Then read Codex vs Claude Code before buying a second Max-class subscription.