GPT-6.1 Terra Is Not Out Yet: What the Leak Shows and Why People Expect It Soon

What the leak actually says

The circulating screenshot shows an evaluation harness configuration. The meaningful lines are:

runs: ${RUNS=5}
model: ${MODEL=gpt-6.1-sol}
judge_model: ${JUDGE_MODEL=gpt-6.1-terra}
judge_reasoning_effort: ${JUDGE_REASONING_EFFORT=medium}
judge_timeout: ${JUDGE_TIMEOUT=10m}
executor: copilot-sdk

Read it as engineering, not marketing. The executor is a released model; the judge is a name that does not exist in any public catalog. An evaluation script that points at a non-callable model produces errors, so the conservative reading is that this identifier resolves inside at least one large customer's environment. The screenshot also states that a batch of evaluation configs moved from GPT-5.6 Terra to GPT-6.1 Terra.

Treat the provenance as unverified. We could not independently confirm the file, the repository or the author's description of it. Everything below is inference built on two solid facts — what OpenAI currently sells, and how the previous generation was structured — plus one leaked artifact.

Why "soon" rather than "someday"

The 6.1 generation has exactly one SKU. GPT-6.1 Sol is described as near-Astra performance at a fraction of Astra's price. A generation rarely stays at a single tier, and the Terra slot — the mid-tier workhorse of the 5.6 family — is conspicuously empty.

Enterprise evaluation tends to lead public launch. Teams migrate traffic only after their internal evals pass, so the harness changes first. A judge model wired into someone's eval pipeline usually trails a public announcement by days to weeks, not quarters.

The naming pattern sets the default expectation. The 5.6 family shipped as Sol (flagship), Terra (balanced workhorse) and Luna (cheap). The community assumes 6.1 will fill the same shape, with Sol opening and Terra following.

The obvious counterweight: identifiers inside vendors' configs have surfaced before and then shipped under a different name, later, or not at all. No date, price, context length or benchmark number is public.

What it would likely be, if it ships

Positioned by analogy with GPT-5.6 Terra, GPT-6.1 Terra should land between Sol and Luna: cheaper than the flagship, stronger than the small model. That is the tier for batch evaluation, long-running agents, document processing and high-volume extraction.

The judge-model role is a clue in the same direction. Judges need to be cheap, consistent and stable rather than top-of-leaderboard, which is exactly what a balanced mid-tier model is for. If the plan were to push a new flagship, it would not debut as a grader.

Pricing will almost certainly sit between Sol and Luna. Anyone quoting an exact number today is guessing, so this article does not.

What to do now

If you have deadlines, do not wait. Run gpt-6.1-sol today — the model exists, the routes exist, and the price is roughly one-fifth of Astra's official rate before any discount. Existing GPT-5.6 Terra workloads have no reason to migrate on a rumor.

If you want to catch the launch early, watch two surfaces: the Terra entry appearing in OpenAI's model catalog, and the APIMaster model list, where the card picks up live route prices and discounts once a model is actually sellable.

To get set up: create an APIMaster account, generate an API key, then point your client at https://apimaster.ai/v1 with model: "gpt-6.1-sol". When a Terra tier ships with live routes, switching is a model field change — no integration work.

Checked October 20, 2026. This article is compiled from public sources and one circulating screenshot; it is not an OpenAI announcement. Release timing, pricing and specifications are unknown and will be governed by the official docs.