OpenAI's GPT-5.6 lineup lands on NetRoom: Sol, Terra and Luna
On July 9, 2026 OpenAI opened up the GPT-5.6 family: flagship Sol, balanced Terra and fast Luna. All five versions, Pro modes included, are now on NetRoom.
In short
On July 9, 2026, OpenAI opened public access to the GPT-5.6 series — three models named Sol, Terra and Luna. The family had been running in a limited preview since June 26 for roughly twenty organizations vetted by the US government; after the public release it became available through the API and on OpenRouter. NetRoom has added all five versions: GPT-5.6 Sol, GPT-5.6 Sol Pro, GPT-5.6 Terra, GPT-5.6 Terra Pro and GPT-5.6 Luna. All of them are live in the chat right now.
What the whole series shares
Technically it is one lineup in three sizes, and the core specs are identical across tiers:
- Context window — 1.05M tokens: a mid-size codebase, a thick book or months of correspondence fit into a single request.
- Maximum output — 128K tokens.
- Training data — up to February 16, 2026.
- Six reasoning-effort levels — from none to max — plus a separate Pro mode.
- New in the API: programmatic tool calling, multi-agent support (beta), explicit prompt caching and persisted reasoning — the model can carry its chain of thought across calls instead of starting from scratch.
Alongside the series OpenAI introduced ChatGPT Work, an agent built to carry out whole jobs rather than answer questions. That is a separate OpenAI product and it is not on NetRoom: we plug in the models themselves, so the rest of this post is about them.
Three tiers: which one is for whom
Sol — the flagship for hard problems
GPT-5.6 Sol is the top model of the series, built for heavy tasks and long autonomous sessions: refactoring a large repository, multi-step data analysis, agent work where the model plans and checks itself. On Agents' Last Exam — an evaluation of long professional workflows across 55 fields — Sol sets a new high of 53.6, beating Claude Fable 5 in adaptive reasoning mode by 13.1 points. Sol scores 88.8% on Terminal-Bench 2.1 and 64.6% on SWE-Bench Pro.
Terra — the balanced middle
GPT-5.6 Terra is the mid tier at half the price of Sol. It roughly matches the previous flagship GPT-5.5 and trails Sol by only a few points on benchmarks: 87.4% vs 88.8% on Terminal-Bench 2.1 and 63.4% vs 64.6% on SWE-Bench Pro. Terra has already settled into OpenRouter's category leaderboards: Academia #20, SEO #22, Legal #36, Marketing #40, Finance #50. For most everyday work it is the most sensible price-to-quality pick.
Luna — fast and cheap
GPT-5.6 Luna is the smallest tier: five times cheaper than Sol on input tokens, with the same 1.05M context. Its home turf is high-volume scenarios where speed and per-request cost matter: classification, summarization, support chatbots, text drafts, bulk document processing.
What the Pro versions are
Sol Pro and Terra Pro are not separate models — they are the same Sol and Terra with reasoning.mode=pro switched on. The model thinks longer and spends more tokens on reasoning, but gains noticeably on hard problems: math, architecture decisions, legal analysis, tangled code. Per-token pricing is the same as the base version — the extra cost comes only from longer reasoning chains and response time. The simple rule: start with the base version and switch to Pro when the answers stop being good enough.
Pricing
OpenRouter list prices per 1M tokens:
| Model | Input, $/1M | Output, $/1M |
|---|---|---|
| Sol / Sol Pro | $5.00 | $30.00 |
| Terra / Terra Pro | $2.50 | $15.00 |
| Luna | $1.00 | $6.00 |
NetRoom prices in local currency are listed on each model card in the catalog. Billing is pay-as-you-go: top up your balance and pay only for the tokens you actually use — no OpenAI subscription required.
Practical examples
- Codebase audit. Load an entire repository into Sol's context and ask for bottlenecks and a refactoring plan — 1.05M tokens covers hundreds of files.
- Contracts and documents. Terra Pro walks through a long contract, flags risky clauses and explains them in plain language — Terra's spot in OpenRouter's Legal top is no accident.
- High-volume processing. Luna cheaply churns through thousands of reviews, emails or tickets: classification, summaries, field extraction.
- Research. Sol Pro at maximum reasoning effort — for tasks where every logical link matters: financial models, literature reviews, complex migrations.
The release context
The GPT-5.6 launch happened in two stages, which is news in itself. First came a limited preview starting June 26, open only to organizations vetted by the US government, with access through the OpenAI API and Codex. The public opening on July 9 followed once that review wrapped up. It is the first time a flagship model of this scale has shipped through a government review procedure, and by all signs this will become the norm for future releases. Early reviews are cautiously positive: Sol is described as a very strong model, though on some complex coding tasks reviewers do not yet see a decisive lead over competitors — one more argument for picking the tier that fits the task rather than defaulting to the most expensive one.
Which one to pick: a quick cheat sheet
- Everyday work tasks, documents, texts — Terra.
- Complex analysis, legal and financial reviews — Terra Pro.
- Large codebases, autonomous agent sessions — Sol.
- Maximum quality regardless of response time — Sol Pro.
- High-volume and streaming scenarios, bots, drafts — Luna.
How to try it
All five models are already in the NetRoom chat: open the chat, pick the GPT-5.6 tier you need from the model list and type your prompt. Model pages with specs, examples and current prices live in the /models catalog. If you are unsure where to start, go with Terra: it is the most universal tier of the series. Move up to Sol and the Pro modes when you hit a quality ceiling, and down to Luna when the token bill matters more than the last few benchmark points.