GPT 5.6 Terra — the GPT-5.6 tier that took over the mini slot, its thirty-nine official rows, and what this chat box actually switches on

What is GPT 5.6 Terra?

GPT-5.6 Terra is the middle tier of OpenAI’s GPT-5.6 family, and the official model page describes it in one sentence: it is designed for workloads that balance intelligence and cost, and it roughly corresponds to the mini model tier used in earlier GPT-5 families. The model catalog summarises it the same way, as the GPT-5.6 model that balances intelligence and cost; the family page calls it a lower-cost model with performance competitive with GPT-5.5; and the family guide’s own recommendation is to use it for strong performance at a lower price. The naming scheme is new with this generation: the number identifies the generation, while Sol, Terra and Luna are durable capability tiers that can advance on their own, and the three positions run flagship, balanced and fastest-and-cheapest in that order. OpenAI writes the name GPT-5.6 Terra, while this site’s own label for it — the one the heading and the navigation show — drops the hyphen and reads GPT 5.6 Terra; both appear here, and the model code settles which is meant. Its API model code is gpt-5.6-terra, its default snapshot is the same string with no dated suffix, the documented context window is 1,050,000 tokens, the documented maximum input is 922,000 tokens, the documented maximum output is 128,000 tokens, the documented inputs are text and image, the documented output is text, and the knowledge cutoff is 16 February 2026. Two statements about naming should be separated before anything else. The family alias gpt-5.6 does not point here: the official guide states that the gpt-5.6 alias routes requests to gpt-5.6-sol, and this model is not published with an alias of its own. And there is no gpt-5.6-mini: a request for that model ID returns a not-found page, because the slot that name suggests is the one this model took over. On currency, the model catalog now features GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, and describes GPT-6 Astra as the flagship model for complex reasoning and coding, so this generation has been succeeded; the deprecations page, meanwhile, lists gpt-5.6-terra in neither its upcoming nor its past table, and instead prints it in the replacement column for fifteen older models. Succeeded and deprecated are different states, and both official pages are reported here rather than one of them being picked.

What GPT-5.6 Terra is worth using for

Official GPT-5.6 Terra specifications

Developer

OpenAI

Model ID

gpt-5.6-terra

Official context window

1,050,000 tokens

Official maximum output

128,000 tokens

Official inputs

Text and image

Output

Text

GPT-5.6 Terra and GPT-5.6 Sol side by side

Every value below comes from a single page — the GPT-5.6 family announcement on openai.com, checked on 30 September 2026 — and from a single table family within it. That page carries eleven HTML tables, one per evaluation group, and both the GPT-5.6 Terra and the GPT-5.6 Sol column are present in all eleven. Thirty-nine rows report a value for both models, and all thirty-nine are reproduced here, in the order the official page prints them, so nothing is dropped for going the wrong way and nothing is selected for going the right way. The left column is this page’s model, the middle column is GPT-5.6 Sol, the tier above it; a third GPT-5.5 column exists on the same page and is not reproduced here. Read the direction first, because it is not the direction a reader assumes: Terra wins two of the thirty-nine rows and loses thirty-seven. The two wins are worth naming, since both sit in the self-improvement group and both are wide — NanoGPT at 14.5 against 9.69 and PostTrainBench Lite at 51.5 against 50.3. Among the thirty-seven losses the margins run from a near-tie such as Internal Research Debugging Evaluation at 67.8 against 68.3 up to FrontierMath Tier 4 at 68.3 against 83, ExploitBench at 52.9 against 73.5 and OSWorld 2.0 at 50.2 against 62.6. One row deserves reading twice for a different reason: on the long-context recall test MRCR v2 eight-needle at 512K–1M, Terra scores 72.5, Sol scores 73.8 and GPT-5.5 scores 74, which makes this the one row in the set where the lower tier is below both — and it sits in the same group as the row with the widest gap, GraphWalks BFS 256k f1 at 76.9 against 90.7. Two groups are labelled internal by OpenAI itself, and those rows are the provider’s own evaluations rather than public benchmarks: Management Consulting Tasks and MedChemBench. One conflict inside that same official page is recorded rather than smoothed over: its prose says the generation sets a new high of 53.6 on Agents’ Last Exam, while its own tables print 52.7% for Sol and 50.4% for Terra. This page uses the table values and reports the discrepancy. One further figure travels with the ExploitGym row, where the table prints the six-hour result at 33.7% for Sol and the prose adds that the two-hour figure rose from 15.1% to 24.9%. These are provider evaluations of the models, not measurements of this website, and no score here predicts the quality or speed of your conversation.

How to use GPT 5.6 Terra online

1. Choose the model

   This page opens with GPT 5.6 Terra selected. The model navigation on the left and the selector inside the chat box stay in sync, and switching models keeps the current conversation, so start a new chat when you want a clean context. Three naming facts are worth fixing before you send anything. Two spellings of the name appear on this page and both are correct in their own place: the official model code is gpt-5.6-terra, and the site’s own display label, which is what the navigation and the heading show, omits the hyphen. The family alias gpt-5.6 does not point here — it routes to gpt-5.6-sol — and this model is not published with an alias of its own. And there is no gpt-5.6-mini model ID; the slot that name suggests is the one this model took over.

2. Decide what you expect from reasoning, because this box will not ask

   The family documentation lists six reasoning efforts — none, low, medium, high, xhigh and max — with medium as the default when the setting is omitted, plus a separate pro execution mode available in the Responses API whose supported efforts begin at medium. Persisted reasoning is the part that affects a long session: for this generation the default is to carry earlier turns forward, and reasoning can be reused only within the same model family, so gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna can reuse each other’s reasoning while nothing carries between this family and GPT-5.5. Reasoning tokens are never shown to the caller, still occupy space in the context window, and are billed as output tokens, so their cost lands in whatever follows them. None of that is selectable here: the catalog entry marks reasoning as unsupported, so the provider default is what runs. The family guidance also notes that this generation tends to be more concise than GPT-5.5 by default and exposes answer length as a separate parameter, which this box does not surface either — so state the length you want in the prompt itself.

3. Sign in, send, and check the cost against the official pages

   Sending a message needs an account with available points. The composer reads a reference price from the live model catalog and a finished request reports what it actually cost. Two ceilings are set by this site rather than by the model, so plan around them: replies are requested with at most 4,096 output tokens, and there is no way to raise or inspect the reasoning effort from here. Because the rates quoted on this page are the current standard rates printed by the official pages, re-check them there before relying on a figure for planning, and treat any third-party summary that still quotes a launch price as out of date.

What GPT 5.6 Terra costs, and why this page carries no end date

This model has no end date on it, and that absence is the first thing to establish. The official model page prints three rates for gpt-5.6-terra and two long-context rules and stops there, and the one dated sentence on the pricing page names GPT-5.6 Sol. So the figures below are current standard rates with no window attached, and the only dated price event in this model’s record is a completed one. Both pages were checked on 1 October 2026. At or under 272,000 input tokens the rate is $2.00 per million input tokens, $0.20 per million cached input tokens, $2.50 per million cache-write tokens and $12.00 per million output tokens; above that threshold the same four figures become $4.00, $0.40, $5.00 and $18.00. The stated rule is that prompts with more than 272K input tokens are priced at 2x input and 1.5x output for the full request, so crossing the line reprices the entire request rather than only the excess. Cache writes are billed at 1.25 times the uncached input rate, which is why cache-write tokens get their own column. Batch and Flex are half the standard rate and Fast is twice it, with Fast being the renamed form of what used to be called Priority and the rename dated 30 July 2026; Ultrafast has a row for gpt-6-astra and none for this model. Regional processing for data residency adds 10% for models released on or after 5 March 2026, and FedRAMP endpoints add 10% as well. The dated part is history, not an offer. The family page’s availability section still prints the launch figures for this model — $2.50 per million input tokens and $15 per million output tokens — and carries an update dated 30 July 2026 recording that OpenAI cut the price of GPT-5.6 Terra by 20%. The arithmetic closes on both figures: $2.50 × 0.8 = $2.00 and $15 × 0.8 = $12.00, which is exactly what the model page, the pricing page and the model page’s quick comparison table all print today. So the launch price and the reduction sit side by side on one official page, and neither is a live offer with an end date. A summary that carries the flagship’s November date over to this model has moved it without cause. This site charges none of those figures. It bills points, and the catalog reference price behind this page is 153 points per million input tokens, 918 per million output tokens and 15.3 per million cached input tokens, with no conversion between those and the official dollar rates.

GPT 5.6 Terra FAQ