GPT-6.1 Terra: Capabilities, Pricing, and Practical Use for Your Workflows

Terra at a glance: where it shines

Terra is described as fast and efficient, especially for tasks that blend visual understanding, computer use, and writing. It is designed for everyday agentic work, coding support, and communications tasks that fill a typical workday—such as drafting proposals, interpreting technical drawings, filling out forms, and editing videos. Its pricing is engineered to be accessible for high‑volume automation, helping developers automate routines like invoice processing and data entry into accounting systems. In practical terms, Terra’s positioning makes it a strong candidate when you want a balance between cost and capability for regular, repeatable workloads that involve screen‑based interactions and light to moderate reasoning.

Benchmarks and performance: what the numbers say

In internal benchmarking, Terra shows meaningful cost‑to‑performance advantages across several dimensions. At medium reasoning effort, Terra scores about 15 percentage points higher than GPT‑5.6 Terra on the Agents’ Last Exam, while costing roughly one‑third as much. The model also demonstrates cost efficiency on par with GPT‑6 Astra at only about one‑tenth of Terra’s price per token, illustrating strong value for medium‑difficulty tasks. At maximum effort, Terra becomes notably cost‑efficient: it is over three times as token efficient as Claude Opus 5.5 and costs about a quarter of its per‑token price, yielding an approximately nine‑fold reduction in cost per task. Importantly, Terra completes tasks more quickly—over three times faster—than both GPT‑6 Astra and GPT‑6.1 Sol in these comparisons.

Additional internal benchmarks corroborate Terra’s strengths. On MMMU‑Pro, a test of visual reasoning, Terra outperforms GPT‑5.6 Terra at every reasoning setting and approaches the performance of GPT‑6 Astra and GPT‑6.1 Sol at a much lower cost. OSWorld 2.1 tests of desktop‑app automation show Terra surpassing 5.6 Terra’s best score at medium reasoning effort, again at a fraction of the cost.

These results matter because they indicate Terra’s niche: cost‑effective performance in tasks that require visual understanding, on‑screen actions, and practical writing. They also set expectations that Terra excels in routine, high‑volume workflows but may not always match the frontier performance of Sol or Astra on the most demanding coding or multi‑domain reasoning tasks.

Why this matters for you: if your workflows involve lots of form processing, invoice extraction, or rapid on‑screen decision making, Terra’s cost per task and throughput can translate into meaningful savings without sacrificing reliability. If your work hinges on highly complex coding or frontier algorithms, you may prefer higher‑tier options like Sol or Astra for certain tasks, even if Terra remains compelling for many routine use cases.

Safety, alignment, and factual improvements

Terra marks progress in factual accuracy and alignment within OpenAI’s safety framework. In internal evaluations focused on reducing hallucinations and factual errors, Terra’s error rate at low reasoning effort dropped from 53.6% (5.6 Terra) to 16.7% (6.1 Terra), a substantial improvement. At maximum effort, the error rate fell by 18 percentage points to 4.3%, approaching the levels of GPT‑6 Astra and GPT‑6.1 Sol. In software engineering tasks, Terra significantly improves coding performance: Terminal‑Bench 4.0 shows Terra at maximum reasoning effort solving 50.3% of tasks, more than double the 21.5% published for 5.6 Terra. For the more advanced coding benchmark Terminal‑Bench‑Science 0.1, Terra closes the gap with Sol and approaches Opus 5.5 on cost at max reasoning.

On safety and alignment, Terra is described as staying within scope more reliably and disclosing potential limitations to users, with strong resistance to prompt injections in tests designed to probe malicious instructions. These safeguards enhance trust, particularly for developers integrating Terra into agentic workflows where transparency and reliability matter.

Pricing, availability, and deployment options

Terra is available to Plus, Pro, Enterprise, and Edu users within ChatGPT Work and Codex, with developers able to access it through the OpenAI API under the slug gpt-6.1-terra. The standard API pricing is $1 per million input tokens and $5 per million output tokens. Cache reads are priced at $0.05 per million tokens, and cache writes at $1.25 per million tokens. Web search incurs a separate $10 per 1,000 calls. Terra supports a large context window of 1,050,000 tokens and up to 128,000 completion tokens.

OpenRouter provides routing across multiple providers with modes designed to balance cost and speed. Balanced prioritizes a mix of price and performance, Nitro emphasizes speed, and Floor targets the lowest cost. EU in‑region routing allows requests to be decrypted in the European Union and routed only to European provider endpoints, with a fails‑closed approach if no EU endpoint is available. In regions outside the EU, regional routing and failover are managed to maintain availability and throughput. Throughput is measured at 168 tokens per second (P50) with a latency of about 1.27 seconds (P50) at best across providers, and uptime around 99.99% over a three‑day window, with near‑real‑time routing adjustments as needed.

Key data points from the model’s operational profile include a 1,050,000 token context and the ability to handle up to 128,000 completion tokens, together with tool‑calling support and structured outputs via a JSON schema. Terra is rated to work with a variety of inputs and outputs, including PDFs, images, and text, and supports multiple endpoints for live sessions, chat completions, and batch processing. Providers actively competing in this space include OpenAI and Azure, with OpenRouter routing requests to the best available provider and offering explicit controls to pin or exclude providers as needed.

Specific third‑party benchmarks and internal analyses frequently cited in industry discussions show Terra’s strengths in cost efficiency, with a notable comparative advantage at medium and low reasoning effort levels. These benchmarks also highlight where Terra is not Pareto‑optimal, particularly for the most demanding coding tasks where Sol or Astra can outperform Terra on certain metrics. When planning deployment, consider Terra as a cost‑efficient option for routine, high‑volume agentic tasks and writing work, rather than a one‑size‑fits‑all replacement for higher‑tier models.

Decision framework: choosing Terra, Luna, Sol, or Astra for common tasks

To translate Terra’s strengths into a practical decision guide, use this framework when selecting among Terra, Luna, Sol, and Astra for typical workflows. The guidance below reflects Terra’s cost advantages, its efficiency for routine tasks, and where higher‑tier models tend to outperform Terra on more demanding work.

In practice, use Terra when you need a reliable middle ground: strong visual understanding, practical on‑screen use, and writing with good cost performance. If you face frequent complex coding demands or ambitious software projects, consider Sol or Astra for their frontier capabilities, even if Terra would be cheaper per token. For the most budget‑conscious, Luna remains the baseline option for simple, high‑volume tasks.

Concrete use cases and workflows

Concrete workflows cited in industry discussions illustrate Terra’s practical applications. One common use case is automated document extraction: Terra processes PDF invoices and writes the extracted data to a database, enabling faster accounting workflows. Another recurring theme is on‑screen task automation and reviews: Terra is leveraged for PR reviews in code repositories, translation or SQL translation tasks, and help‑desk ticket generation, delivering cost and speed advantages in day‑to‑day operations. For more complex, long‑form analysis or data‑heavy research tasks, Terra is often paired with higher‑tier models to balance speed, cost, and depth of reasoning. Finally, Terra demonstrates strength in scenarios requiring visual reasoning and on‑screen manipulation, such as turning architectural drawings into usable digital artifacts or logging frame‑by‑frame content for a restoration project.

Two practical workflows you might adopt today include: (1) automated invoice processing where Terra reads incoming invoices, extracts key fields, and writes them into an accounting system, and (2) staff support automation where Terra translates a customer brief and related documents into a structured design blueprint or ticket and routes it to the appropriate team for execution. In both cases, Terra’s balance of speed and cost makes it competitive for sustained use in production environments.

Caching, context, and prompt management: actionable guidance

Prompt caching can significantly reduce repetitive prompt costs, but awareness of how caching works is essential to avoid surprises. Terra caches prompts automatically for content blocks of 1,024 tokens or more, with a cache write charged on each cached prompt. To disable caching for a particular request, set prompt_cache_options.mode to “explicit” at the request root and avoid marking any content block with a prompt_cache_breakpoint. In explicit mode, OpenAI does not apply its own breakpoints, so a request with no cacheable prefix is billed at the plain input rate. Prompts longer than 272,000 tokens are priced at 2× the input tokens plus cache costs and 1.5× the output tokens for the full request. Batch and Flex pricing are 50% lower than Standard pricing, and regional processing adds a 10% premium where available. If you’re working with long contexts, keep these modifiers in mind and consider using explicit mode when your prompts vary significantly between calls to avoid paying unnecessary cache writes.

From a deployment perspective, a simple rule of thumb is: if your prompts vary a lot between requests, disable caching to avoid cache write charges; if prompts are highly repetitive across many calls, enabled caching can save substantial cost. For workloads with very large context windows, remember that pricing scales with context length, and batch/flex options can further reduce the per‑token cost when appropriate.

Coding capabilities and comparisons: Terra versus competitors

Terra’s coding capabilities show a clear strength in agentic tasks and visual reasoning, with caveats. Public benchmarking suggests Terra’s coding performance sits between Luna and Sol depending on the exact task and reasoning level. In available internal scorings, a direct comparison indicates that, for some coding tasks, 6.1 Sol and Astra can outperform Terra at higher reasoning levels or in more logic‑intensive scenarios. Third‑party analyses report that Terra’s Coding Agent Index sits lower than Sol (TerrA 48 vs Sol 54) but higher than Luna (31), indicating Terra offers a solid middle ground for coding tasks with notable cost efficiency. In Terminal‑Bench 4.0, at max reasoning effort, Terra achieves a 50.3% task completion rate, more than double 5.6 Terra’s 21.5%, while Astra and Sol still hold their own on many complex tasks. For scientific coding tasks (Terminal‑Bench‑Science 0.1), Terra approaches Opus 5.5 on cost while maintaining competitive performance, though Sol and Astra can still lead on very demanding problems. In practical terms, Terra’s coding strengths lie in agentic tasks and on‑screen workflow automation, while highly complex pure code optimization or very deep logic puzzles may be better served by Sol or Astra, depending on cost and throughput goals.

When weighing Terra against Claude Sonnet 5.5 or DeepSeek Flash, the landscape depends on the task. Across benchmarks, some coders find Terra sufficiently capable for routine coding assistance and automation at a fraction of the cost, while others prefer dedicated coding engines for very large codebases or highly intricate logic tasks. If coding is a primary driver for your use case, consider Sol or Astra for hard coding and use Terra to fill gaps where cost and speed are decisive factors.

Limitations and caveats: where Terra may not be the best fit

Readers should understand Terra’s positioning as a mid‑tier option with strong cost efficiency for routine tasks but not guaranteed to outperform higher‑tier models on all dimensions. Some user feedback points to Terra as an “unhappy middle” for certain coding scenarios or highly specialized workloads where Sol or Astra might offer better price‑performance. Versioning and naming can also introduce confusion: Terra’s release cadence and cross‑model comparisons can create uncertainty about which variant to pick for a given task. Always corroborate open guidance with your own internal evals when cost and accuracy are both critical. If you rely on highly complex or creative workloads, testing Terra alongside higher‑tier options remains a prudent approach to ensure you achieve the right balance of cost, speed, and capability for your use case.

Concrete usage notes for deployment and optimization

For teams prioritizing cost per task and predictable performance, Terra’s pricing structure—$1.00 per million input tokens, $5.00 per million output tokens, $1.25 per million cache writes, and $0.05 per million cache reads—offers a transparent baseline. The 1,050,000‑token context window and up to 128,000 completion tokens support sizable interactions, while tool calling and structured outputs enable automation workflows. When routing via OpenRouter, choose a mode that matches your priorities: Balanced for a mix of cost and speed, Nitro for speed, or Floor for the lowest cost. If you operate in the EU, enable in‑region routing to keep data local and reduce latency, with automatic failover to other endpoints if needed. In practice, for forms processing and repetitive data tasks, Terra often delivers the best return on investment among the mid‑tier options, provided you select a reasoning level that matches the workload and disable caching for highly variable prompts when necessary.

Use‑case examples you can start with

Two practical workflows you can explore right away include:

Automated document extraction: Terra processes PDFs such as invoices and writes the extracted fields to a database, enabling streamlined accounting workflows without manual data entry.

Help‑desk ticket generation: Terra translates customer briefs and related documents into structured tickets, then routes or escalates as appropriate, reducing cycle time for issue handling.

Final takeaway

GPT‑6.1 Terra offers a compelling combination of speed, on‑screen utility, and cost efficiency for everyday, high‑volume tasks. It is well suited for routine document processing, form handling, and on‑screen workflows, where its medium‑effort cost efficiency translates into meaningful savings. For highly demanding coding workloads or frontier-style tasks, Sol or Astra often deliver stronger performance, albeit at a higher price. Terra sits in the middle, giving you a versatile option that scales well for many common tasks while keeping a careful eye on cost and deployment practicality.