# GPT-5.6 Terra

GPT-5.6 Terra is an AI model released by OpenAI on Jun 26 2026. It has a 1.05M context window. Tracked results include 54% on BullshitBench v2, 92.9% on GPQA Diamond and 69.6% on DeepSWE 1.1.

## Facts

| Field | Value |
| --- | --- |
| Model | GPT-5.6 Terra |
| Developer | OpenAI |
| Release date | Friday, Jun 26 2026 |
| Licensing | Closed |
| Context window | 1.05M |

## API pricing

All rates in USD per 1,000,000 tokens, pay-as-you-go.

| Tier | Input | Cached input | Cache write | Output |
| --- | --- | --- | --- | --- |
| Up to 272K input tokens | $2.00 | $0.20 | $2.50 | $12.00 |
| Over 272K input tokens | $4.00 | $0.40 | $5.00 | $18.00 |

Verified October 19, 2026 against the first-party source: https://developers.openai.com/api/docs/models/gpt-5.6-terra

## Tracked benchmark scores

| Benchmark | Score | Source | What it measures |
| --- | --- | --- | --- |
| BullshitBench v2 | 54% | [BullshitBench](https://github.com/petergpt/bullshit-benchmark) | Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better. |
| Gray Swan IPI (k = 1) | 5.4% | Lab | Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. Gray Swan's indirect prompt injection benchmark measures how often such an attack succeeds when the attacker gets a single try. Lower is better. |
| Gray Swan IPI (k = 10) | 26% | Lab | Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. This variant gives the attacker 10 tries and counts an attack as successful if any of them works. Lower is better. |
| Gray Swan IPI (k = 15) | 30.4% | Lab | Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. This variant gives the attacker 15 tries and counts an attack as successful if any of them works. Lower is better. |
| SWE-Bench Pro | 63.4% | Lab | Can the AI fix real bugs in real software? It's handed actual problems from open-source projects and has to write code that genuinely solves them. Higher is better. |
| CursorBench 4.0 | 41.3% | [CursorBench](https://cursor.com/cursorbench), retrieved 2026-09-28 | Cursor's own test of coding agents on ambiguous, multi-file tasks taken from real Cursor sessions — editing, refactoring, investigating a codebase, understanding what the user meant, managing jobs and following a design. Cursor runs each model at several reasoning efforts; each release here carries the score of its best listed effort. Scores aren't comparable with earlier CursorBench versions. Higher is better. |
| DeepSWE 1.1 | 69.6% | Lab | Artificial Analysis' independent test of deep, agentic software-engineering work — the AI has to plan and carry out substantial coding tasks end to end. (Version 1.1 of the test.) Higher is better. |
| BenchCAD (without tools) | 62.3% | Lab | geometric overlap of generated CAD parts without tool access |
| BenchCAD (with tools) | 78.2% | Lab | geometric overlap of generated CAD parts with tool access |
| Frontier-Bench v0.1 | 20.8% | Lab | A hard, ever-evolving set of real computer tasks — coding, system administration, data work, and more — that an AI agent has to complete on its own. Run by the Harbor / Laude Institute team as the successor to Terminal-Bench (v0.1 is the first release of the task set). The score is the share of tasks solved. Higher is better. |
| Terminal-Bench 4.0 | 21.52% | [Terminal-Bench](https://github.com/harbor-framework/terminal-bench), retrieved 2026-08-30 | Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Version 4.0 recalibrated how much time, CPU and memory each task gets, removed eight tasks and fixed nineteen, so fewer runs fail for reasons that have nothing to do with the model. Scores are not comparable with earlier versions. Higher is better. |
| Terminal-Bench 2.1 | 87.4% | Lab | Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Higher is better. |
| Toolathlon | 53.1% | Lab | Tests how well the AI uses everyday real-world tools and apps to get things done. Higher is better. |
| BrowseComp | 87.5% | [BenchLM](https://benchlm.ai), retrieved 2026-08-24 | Can the AI browse the web and track down hard-to-find answers? Higher is better. |
| ARC-AGI-3 | 0.8% | Lab | The third generation of the ARC-AGI series: instead of static puzzles, the AI is dropped into small interactive game-like environments it has never seen and has to figure out the rules and solve them on its own. Higher is better. |
| ARC-AGI-2 | 83.9% | [BenchLM](https://benchlm.ai), retrieved 2026-08-24 | Puzzle-style tests of abstract reasoning and pattern-finding — the kind of thing people find easy but AIs often struggle with. Higher is better. |
| FrontierMath (Tier 4 (v2)) | 68.3% | Lab | The rebuilt edition of FrontierMath's hardest tier — research-level maths of the kind professional mathematicians work on. It is a different question set from the first Tier 4, and scores on it run far higher, so read the two as separate tests rather than progress. Higher is better. |
| GeneBench-Pro | 23.3% | Lab | Real genomics and biomedical analyses done end to end: the AI gets a messy dataset and a question, and has to work through the chain of statistical decisions to a verifiable answer — the job a computational biologist does before a research or clinical decision gets made. Higher is better. |
| MedChemBench (Internal) | 35% | Lab | OpenAI's own drug-discovery test: reading chemical structures, predicting how potent or toxic a compound will be, choosing between candidate molecules, and planning a synthesis route — the everyday judgement calls of a medicinal chemist. Higher is better. |
| LifeSciBench | 56% | Lab | performance on research tasks graded by life sciences experts |
| FrontierMath (Tiers 1–3 (v2)) | 84.9% | Lab | research mathematics on the rebuilt second edition of tiers 1–3 |
| GPQA Diamond | 92.9% | Lab | Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. |
| Agent's Last Exam (pass@1) | 50.4% | Lab | A hard set of desktop and operating-system tasks an AI agent has to finish by looking at the screen and working the machine itself. The score is the share it passes outright — partial credit does not count. Higher is better. |
| AutomationBench | 15.2% | Lab | Tests whether the AI can run real multi-step business workflows — the kind of end-to-end office processes companies want to automate — from start to finish. Higher is better. |
| HealthBench Professional (OpenAI grading, length-adjusted) | 57.7% | Lab | official OpenAI grading with the response-length penalty applied |
| GDP.PDF | 24.7% | Lab | Real professional PDFs — filings, reports, technical documents — with questions an expert in that field would ask. Tests whether the AI reads the page as a document, layout and figures included, rather than as loose text. Higher is better. |
| MMMU-Pro (with tools) | 82% | Lab | multimodal understanding and reasoning with access to tools |
| MMMU-Pro | 80.7% | Lab | A tougher version of MMMU — college-level questions that mix images, diagrams, and text together. Higher is better. |
| MRCR v2 (8-needle) (256K–512K) | 89.6% | Lab | mean match ratio for eight needles within the indicated context band |
| MRCR v2 (8-needle) (512K–1M) | 72.5% | Lab | mean match ratio for eight needles within the indicated context band |
| threejseval | 1429 | [threejseval](https://threejseval.com/ranking) | Every model gets the same prompt — "the Eiffel Tower", "a glass fishbowl", "a robot arm picking toys into a box" — and builds a 3D scene in Three.js. Real people then see two scenes side by side, names hidden, and vote for the one they prefer. The votes become a chess-style Elo rating on threejseval.com, averaged across all the prompts. It measures whether the scene looks and moves right to a human eye, not whether the code passes a test. Higher is better. |

## About GPT-5.6 Terra

GPT-5.6 Terra was announced in the June 2026 GPT-5.6 preview as the balanced model for everyday work, between flagship Sol and the lower-cost Luna tier. OpenAI positioned it around capability and efficiency for frequent professional tasks. Initial access was limited to trusted partners, with general availability coming later for the family.

Its API offered a 1,050,000-token context window and configurable reasoning, preserving substantial input capacity in the middle tier. Terra's role roughly corresponded to the mini tier of earlier GPT-5 releases. GPT-6.1 Terra followed in October 2026 with an emphasis on visual reasoning, computer use and writing; that later release was separate from the original GPT-5.6 preview and rollout.

## Questions and answers

### When was GPT-5.6 Terra released?

GPT-5.6 Terra was released by OpenAI on Friday, Jun 26 2026.

### Who made GPT-5.6 Terra?

GPT-5.6 Terra was built by OpenAI. Creators of ChatGPT and the GPT series of models. Pioneered large-scale language model research.

### How much does GPT-5.6 Terra cost?

GPT-5.6 Terra costs $2.00 per million input tokens and $12.00 per million output tokens through the OpenAI API. Cached input is $0.20 per million tokens. Those are the rates for the “Up to 272K input tokens” tier; 1 other pricing tier is published for this model. Rates are pay-as-you-go API prices verified against OpenAI's published pricing on October 19, 2026.

### What benchmark scores did GPT-5.6 Terra get?

GPT-5.6 Terra reports 31 tracked benchmark scores — BullshitBench v2: 54%; Gray Swan IPI (k = 1): 5.4%; Gray Swan IPI (k = 10): 26%; Gray Swan IPI (k = 15): 30.4%; SWE-Bench Pro: 63.4%; CursorBench 4.0: 41.3%; DeepSWE 1.1: 69.6%; BenchCAD (without tools): 62.3%; BenchCAD (with tools): 78.2%; Frontier-Bench v0.1: 20.8%; Terminal-Bench 4.0: 21.52%; Terminal-Bench 2.1: 87.4%; Toolathlon: 53.1%; BrowseComp: 87.5%; ARC-AGI-3: 0.8%; ARC-AGI-2: 83.9%; FrontierMath (Tier 4 (v2)): 68.3%; GeneBench-Pro: 23.3%; MedChemBench (Internal): 35%; LifeSciBench: 56%; FrontierMath (Tiers 1–3 (v2)): 84.9%; GPQA Diamond: 92.9%; Agent's Last Exam (pass@1): 50.4%; AutomationBench: 15.2%; HealthBench Professional (OpenAI grading, length-adjusted): 57.7%; GDP.PDF: 24.7%; MMMU-Pro (with tools): 82%; MMMU-Pro: 80.7%; MRCR v2 (8-needle) (256K–512K): 89.6%; MRCR v2 (8-needle) (512K–1M): 72.5%; threejseval: 1429. Tracked scores may come from lab reports or independent benchmarks; source details accompany the benchmark data.

### What is the context window of GPT-5.6 Terra?

GPT-5.6 Terra has a context window of 1.05M. Providers may specify separate input and output token limits.

### Is GPT-5.6 Terra open source?

No. GPT-5.6 Terra is a closed model. The weights are not published — it is available only through the provider's own API, apps, or partner platforms.

### What came before and after GPT-5.6 Terra?

OpenAI's previous tracked release was GPT-5.6 Sol on Jun 26 2026. It was followed by GPT-5.6 Luna on Jun 26 2026.


---

Canonical page: https://aireleasetracker.com/model/openai/gpt-5.6-terra
Site index: https://aireleasetracker.com/llms.txt
Source: AI Release Tracker (https://aireleasetracker.com). Most benchmark scores come from lab launch material; gathered results identify the leaderboard that published them.