GPT-6.1 Terra on TopxAI
· openai, models, api, topxai
Half the price of GPT-6.1 Sol, image input, a 1.05M-token context and reasoning you can switch off: prices, settings and a Responses example.
GPT-6.1 Terra is now on TopxAI's OpenAI shared and official routes and on Auto route, as gpt-6.1-terra. OpenAI aims it at visual reasoning, operating application screens and writing. Per token it costs half of GPT-6.1 Sol, it reads text and images, and unlike Sol it can run with reasoning turned off. Sol and GPT-6 Astra stay in the catalogue; nothing is retired.
Call it
Keep the TopxAI base URL and your key. A route key without a model restriction works as it is. If the key has an allowed-model list, add gpt-6.1-terra to it on the API keys page.
curl https://ai.topxea.com/v1/responses \
-H "Authorization: Bearer $TOPXAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"gpt-6.1-terra","input":"Rewrite in plain English: the deliverable is contingent on stakeholder alignment.","reasoning":{"effort":"none"}}'OpenAI serves Terra's tool calls only on Responses. TopxAI therefore sends Chat Completions and Anthropic-compatible requests for Terra to Responses upstream and converts the answer back, streamed tool calls included, so a client that speaks Chat Completions keeps working unchanged.
Prices
Default USD rates per million tokens, checked against OpenAI's model page on October 30, 2026:
| Category | Shared | Official | OpenAI Standard |
|---|---|---|---|
| Input | $0.50 | $0.90 | $1 |
| Output | $2.50 | $4.50 | $5 |
| Cache read | $0.025 | $0.045 | $0.05 |
| Cache write | $0.625 | $1.125 | $1.25 |
The shared route is 50% of OpenAI's Standard price and the official route 90%, as for the other GPT models. Once a request's input passes 272,000 tokens, cached input included, the whole request moves to the long-context tier: on the shared route that is $1 input, $3.75 output, $0.05 cache read and $1.25 cache write. The model page lists both tiers for both routes.
Cache writes are sold at OpenAI's default 30-minute retention. A 12-hour ttl is not forwarded, whether it sits in prompt_cache_options or on one prompt_cache_breakpoint in input; the second case was forwarded until this release, for every OpenAI model. How requests are billed has the details.
Reasoning
The context window is 1,050,000 tokens, up to 922,000 of them input, and a reply can run to 128,000 tokens. Terra accepts none, low, medium (the default), high, xhigh and max.
With none the model answers without reasoning tokens, which suits rewrites, extraction and short replies. It is also the only setting under which temperature and top_p reach OpenAI; at any other level TopxAI removes them, because OpenAI does not support them there. Log probabilities are not available at any level: the upstream rejects top_logprobs and the logprobs include, so TopxAI removes those too. An Anthropic-style request with thinking disabled is sent as none. The legacy minimal level, which Terra does not have, becomes low.
For Codex, set model = "gpt-6.1-terra" in ~/.codex/config.toml; the Codex guide covers the rest of the setup.
Models: gpt-6.1-terra · Models & pricing
Related
- GPT-6.1 Terra API pricing
- Moving to GPT-6.1 Sol on TopxAI
- Codex CLI custom model provider: model_providers in config.toml for TopxAI
- How requests are billed
- Moving to deepseek-flash-Abliterated on TopxAI
- Kimi K3.1 Pro replaces Kimi K3 on TopxAI
Other languages: 简体中文