GPT-6.1 Terra lands in the API: the model card, official pricing, and the GPT-6 family price table

GPT-6.1 Terra is OpenAI's "fast, efficient model for visual reasoning, precise computer use, and clear, natural writing" — model ID gpt-6.1-terra, announced for the API in the official changelog on 2026-10-26. Until now this site could only mark its model ID, context, and absolute pricing as unverified (the official docs domain had been returning 403 to us); on 2026-11-01 developers.openai.com became reachable again, and this article completes those parameters line by line from the official model card and pricing page — and, along the way, assembles the official GPT-6 family price table in full for the first time.

1. The launch timeline: one model, three same-day sources

Three independent official channels carried GPT-6.1 Terra on 2026-10-26:

• The official news RSS entry Introducing GPT-6.1 Terra, tagline Bringing fast and capable computer use to everyone (the article page returns 403 to this site, so detail stays at the RSS boundary).
• The GitHub changelog announcing its GA in GitHub Copilot (10 model-picker entry points, available to Pro/Pro+/Max/Business/Enterprise) — the product side is covered in GPT-6.1 Terra announced and live in GitHub Copilot: the everyday agentic coding tier.
• The official API changelog Oct 26 entry (fetched 2026-11-01): Released GPT-6.1 Terra, a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing — tagged Model: gpt-6.1-terra plus the API: v1/responses and API: v1/chat/completions labels.

The SDK type layers followed the same day (openai-python v3.39.0 #4180 / openai-node v7.45.0 #3054); Codex arrived a day later — 0.165.0 (10-27) added Terra to the model catalog, and 0.166.0 (10-29) extended catalog coverage to Vertex AI and Amazon Bedrock (#57293).

2. The model card: context, modalities, effort, and endpoints

Key parameters from the official model card (developers.openai.com/api/docs/models/gpt-6.1-terra, fetched 2026-11-01):

The card's use-case examples after the positioning line: matching product photos to catalog entries, interpreting technical drawings, acting on application screens, and drafting or revising text. reasoning.effort spans all six levels with medium as default — note that Terra supports none, while GPT-6 astra does not (the astra card starts at low). On tools, the card states plainly: Tool calling requires the Responses API.

The endpoint support matrix (card wording): Responses and Chat Completions both Supported, Batch Supported; Live, Realtime, Realtime translation and transcription, fine-tuning, image generation and edits, speech, transcription, and Moderation all Not supported. Ten supported tools: web_search / file_search / image_generation / code_interpreter / hosted_shell / apply_patch / skills / computer_use / mcp / tool_search.

A minimal call (Responses API — the field is input):

from openai import OpenAI

client = OpenAI()
resp = client.responses.create(
    model="gpt-6.1-terra",
    input="Match this product photo to its catalog entry",
    reasoning={"effort": "low"},
)
print(resp.output_text)

3. Pricing: short context, long context, and caching

The official pricing page (fetched 2026-11-01), consistent with the model card:

Prompts over 272K input tokens bill at long-context rates for the full request (card wording: priced at 2x input and cache rates and 1.5x output for the full request) — 2x input and cache, 1.5x output. Three service-tier variants: Batch and Flex are both 50% of standard; regional processing (data residency) endpoints add 10%; Fast mode is 2x standard ($2/$10 short context).

The 5% cached-read discount deserves a callout: the prompt caching guide states the 0.05x tier officially belongs to exactly two models, GPT-6.1 Sol and GPT-6.1 Terra — every other model reads cache at 0.1x of input. The 12-hour cache tier is the capability added by the 2026-10-20 changelog entry (writes at 2.5x the input price) — this site covers it in OpenAI's October changelog, backfilled: GPT-6 Rosalind, ChatGPT subscription API credits, and 12-hour prompt caching.

Rate limits follow the three-tier usage system introduced on 2026-10-06 (model card limits table):

4. In the GPT-6 family: the official price table, complete at last

Short-context standard prices from the pricing page (per 1M tokens, fetched 2026-11-01), joined with the release wording from the September and October changelog backfill:

Two verification notes: first, the 2026-09-29 official RSS line that GPT-6.1 Sol's standard API input/output token prices are one fifth of Astra's matches the price table exactly ($2/$10 against $10/$50); second, the 10-26 GitHub announcement's lower cost than GPT-6.1 Sol for Terra corresponds to exactly half ($1/$5 against $2/$10). Both previously unverified relative-pricing claims now have absolute official prices behind them.

Current GPT-5.6 prices from the same page: Sol $4/$20 (the promotional price, officially guaranteed at least through 2026-11-21), Terra $2/$12, Luna $0.20/$1.20 — Terra and Luna reflect the 2026-07-30 price cuts (-20% and -80%); this site's selection guides have been corrected accordingly.

5. Relation to the Codex-side 500K window

One number invites confusion: the Codex CLI bundled catalog models.json (rust-v0.165.0, verified by this site) records context_window 500000 / max_context_window 872000 for gpt-6.1-terra, while the API model card says 1,050,000 / 922,000. The two do not contradict each other and are not interchangeable — the Codex figures are runtime window settings inside Codex (the GPT-5.6 tiers sit at the same 272K/872K class there), while the model card figures govern API-side limits and billing. Ask which surface you are running on, then pick the table.

6. Common mistakes and troubleshooting

• Treating Codex's 500K as the API limit: the API side follows the model card's 1.05M/922K/128K; input over 272K triggers long-context multipliers, so budget at 2x input and 1.5x output.
• Wiring tools over Chat Completions: the card says tool calling requires the Responses API; Chat Completions is for plain text output only.
• Assuming effort=none is unavailable: Terra spans all six levels (none included, medium default); applying astra's restriction to Terra needlessly spends reasoning tokens.
• Budgeting cache writes at the 30-minute rate while actually using the 12-hour tier: 12-hour writes cost 2.5x the input price ($2.50/MTok), double the 30-minute tier; in a request that mixes tiers, each prefix token bills at the longest lifetime covering it.
• Sending audio or video to Terra: modalities are text and image in, text out; audio pipelines belong to the GPT-Live / Realtime family.

7. Next steps

• The product-side announcement and GitHub Copilot GA: GPT-6.1 Terra announced and live in GitHub Copilot: the everyday agentic coding tier.
• Type-layer and product context for the rest of the family: The GPT-6 family grows: gpt-6-sol, gpt-6-luna, gpt-6.1-sol land in the official SDKs.
• 12-hour caching, subscription API credits, and Rosalind: OpenAI's October changelog, backfilled: GPT-6 Rosalind, ChatGPT subscription API credits, and 12-hour prompt caching.
• The full selection table and decision tree: GPT-5.6 Model Selection Guide (2026): Sol, Terra, or Luna?.