GPT-6.1 Terra vs GPT-5.6 Terra: Price, Context Window and API Changes

The short version

• GPT-6.1 Terra costs $1 for input and $5 for output per million tokens, versus GPT-5.6 Terra’s $2 and $12.
• Both OpenAI models accept text and images, share a 1,050,000-token context window and allow up to 128,000 output tokens.
• The reasoning-effort choices stay the same, with medium as the default.
• GPT-6.1 Terra requires the Responses API for tool calls. Chat Completions remains available for requests without tools.

GPT-6.1 Terra costs half of GPT-5.6 Terra’s input price and well under half its output price: $1 versus $2 for input, and $5 versus $12 for output, per 1 million tokens. Both retain a 1,050,000-token context window, a 128,000-token maximum output and the same reasoning-effort choices, with medium as the default. The required migration change concerns tool calls: GPT-6.1 Terra needs the Responses API for those requests.

How do GPT-6.1 Terra and GPT-5.6 Terra compare on price and limits?

GPT-6.1 Terra lowers token prices while keeping GPT-5.6 Terra’s context and output limits. These are Standard-tier prices per 1 million tokens, read on October 28, 2026.

GPT-5.6 Terra vs GPT-6.1 Terra at a glance

Switching models does not increase how much context a request can hold or how much output it can return. The price difference applies to ordinary input, cached input, cache writes and output.

How did GPT-6.1 Terra compare with GPT-5.6 Terra in Codex tests?

GPT-6.1 Terra completed 9 of 9 runs fully correctly on each task set; GPT-5.6 Terra completed 8 of 9 on the easy set and 6 of 9 on the hard set.

We ran it: GPT-6.1 Terra vs GPT-5.6 Terra in Codex

On October 28, 2026, we ran both models at medium reasoning effort on a Linux server using OpenAI’s Codex CLI 0.159.0 in non-interactive mode through codex exec. Each task ran three times per model across three rounds, with runs executed individually in fresh folders. We used a ChatGPT account sign-in and paid nothing per token; the dollar figures are calculated API Standard equivalents for the tokens used.

• Easy kit: a Python fix checked by 8 tests, 10 questions about our mounting-plate drawing, and a supplier form containing 15 fields extracted from an email.
• Hard kit: a Python package containing 7 bugs, checked by 16 visible tests and 8 hidden tests; 12 questions about our bracket drawing, including 4 multi-step calculations; and a supplier document task containing 20 fields, using a 5-message email thread and a CSV price list.

Each round contained one attempt at each task; the round figures are medians across the three rounds. Codex reported no cache-write tokens, so the calculation priced uncached input using the 30-minute cache-write rate, 1.25 times ordinary input.

Results across both Codex task sets

GPT-5.6 Terra’s input price was 2 times GPT-6.1 Terra’s and its output price was 2.4 times as much; it also consumed more input tokens and more than twice as many output tokens per round, making its round cost 3.44 times as much on the easy kit and 3.45 times as much on the hard kit. Every incorrect run involved the drawing: its easy-kit miss returned 9,137 mm2 for net area and 296.0 g for mass, against correct answers of 8,999 mm2 and 291.6 g. Each hard-kit drawing run contained at least one incorrect calculation involving net face area, batch mass or counterbore volume; in one run, it counted one counterbore when the drawing contained two.

What code changes are needed to migrate from GPT-5.6 Terra to GPT-6.1 Terra?

Change the model ID and move any Chat Completions requests that call tools to the Responses API. The GPT-6 migration guide also specifies how to carry over reasoning effort and remove incompatible parameters.

• Set your model to gpt-6.1-terra. Preserve the current reasoning effort when it is supported, and compare results using representative images and workflows.
• Send requests involving tools through the Responses API at v1/responses. Requests without tools can continue through Chat Completions at v1/chat/completions.
• Configure effort with reasoning.effort in Responses, or reasoning_effort in Chat Completions.
• For effort settings other than none, omit temperature, top_p and top_logprobs. Also omit logprobs in Chat Completions, and remove message.output_text.logprobs from the Responses include array.
• If an older request uses minimal, begin with low and evaluate the results on representative tasks.
• For GPT-5.6 and subsequent models, configure cache lifetime through prompt_cache_options.ttl, replacing the older prompt_cache_retention setting.

How do caching and long prompts affect GPT-6.1 Terra costs?

Cache reads reduce GPT-6.1 Terra’s input charge, while cache writes and requests exceeding 272K input tokens change the applicable rates. The prompt caching guide describes the lifetime settings and cache boundaries.

• Cache reads cost 0.05 times the ordinary input rate, or $0.05 per 1 million tokens before long-context adjustments.
• A 30-minute cache write costs 1.25 times ordinary input, or $1.25 per 1 million tokens. In Responses, a 12-hour lifetime costs 2.5 times ordinary input, or $2.50 per 1 million tokens.
• Once input exceeds 272K tokens, GPT-6.1 Terra applies double the input and cache rates, plus 1.5 times the output rate, across the entire request. GPT-5.6 Terra’s documented long-prompt multipliers are 2 times input and 1.5 times output.
• A cacheable prefix needs at least 1,024 visible input tokens.
• Each request can create up to four cache writes. Implicit caching consumes one of those writes.
• Both implicit and explicit caching are supported. Explicit-only caching uses prompt_cache_options.mode set to explicit, together with prompt_cache_breakpoint.

What new controls and tools does GPT-6.1 Terra support?

GPT-6.1 Terra supports changing reasoning effort between responses through a configuration_update input item, alongside its supported Responses API tools.

Changing effort between responses

The reasoning guide allows effort changes while retaining the original prompt prefix for caching. These updates follow specific conversation rules:

• Use them in standard mode with a single agent. They can change reasoning effort only.
• Insert the configuration_update item between responses, before the next user message in the input array. Keep the request-level reasoning.effort setting unchanged.
• The chosen effort remains active for subsequent responses until another update replaces it.
• Carry updates forward through previous_response_id, or replay them at their original positions when supplying history yourself.
• The response’s reasoning.effort field still displays the request-level setting, rather than the effort chosen by an update.
• Separate configuration updates with other conversation items; two updates cannot sit directly beside each other.
• Do not combine updates with automatic compaction or automatic truncation. Histories containing these updates are also rejected by /responses/compact.
• For explicit compaction, include a compaction_trigger item in a /responses request. Afterward, insert a fresh configuration_update before the next user message to select the desired effort.

Supported Responses API tools

GPT-6.1 Terra’s model page lists support for these tools through Responses:

• Web search
• File search
• Image generation
• Code interpreter
• Hosted shell
• Apply patch
• Skills
• Computer use
• MCP
• Tool search

The computer use guide specifies computer as the tool type. A computer_call supplies actions in sequence through its actions array; return screenshots in a computer_call_output carrying the matching call_id and a computer_screenshot output.

In Codex and ChatGPT

GPT-6.1 Terra appears in Codex’s model list as the Codex rollout reaches accounts on these ChatGPT plans:

• Plus
• Pro
• Business
• Enterprise
• Edu

Availability depends on your plan, region and workspace settings; see our GPT-6.1 Terra launch coverage.

What OpenAI has not said

• As of October 28, 2026, neither Terra model page lists a dated snapshot ID.
• As of October 28, 2026, GPT-5.6 Terra’s model page gives no retirement date.

FAQs

Is GPT-6.1 Terra more accurate than GPT-5.6 Terra?

In our Codex runs on October 28, 2026, at medium effort, GPT-6.1 Terra got 9 of 9 runs fully right on both task sets. GPT-5.6 Terra got 8 of 9 on the easy kit and 6 of 9 on the hard kit, with every miss involving drawing calculations. Both models passed every hard-kit code test, including the 8 hidden tests, and correctly completed all 20 document fields in every run. We did not measure other GPT-5.6 Terra effort settings, direct API execution or other tasks.

Is GPT-6.1 Terra cheaper than GPT-5.6 Terra?

Yes. At Standard-tier rates read on October 28, 2026, GPT-6.1 Terra costs $1 for input and $5 for output per 1 million tokens, compared with GPT-5.6 Terra’s $2 and $12. Cached input costs $0.05 versus $0.20 per 1 million tokens. Cache writes and long-context adjustments also affect the bill.

Does GPT-6.1 Terra have a bigger context window than GPT-5.6 Terra?

No. Each model has a context capacity of 1,050,000 tokens and can generate up to 128,000 output tokens.

Can I use Chat Completions with GPT-6.1 Terra?

Yes, for requests without tools. GPT-6.1 Terra supports Chat Completions through v1/chat/completions, but tool calls must use the Responses API through v1/responses.

What is GPT-6.1 Terra’s default reasoning effort?

The default is medium. Configure it through reasoning.effort in Responses or reasoning_effort in Chat Completions.

Related on NerdsChalk

• GPT-6.1 Terra in Codex: Low vs Medium vs High vs Xhigh
• GPT-6.1 Terra vs Claude Haiku 4.5: Price, Speed and Accuracy Tested
• GPT-6.1 Terra vs Claude Sonnet 5.5: Price, Speed and Accuracy Tested
• GPT-6.1 Terra vs GPT-6.1 Sol vs GPT-6 Luna: ChatGPT Models’ Codex Task Costs
• GPT-6.1 Terra Brings Screen Actions Into Its Visual Toolkit