OpenAI shipped GPT-6.1 Terra this week, and Anthropic's Claude 5.5 family is complete with Haiku 5.5. All six new models are callable on the withConflux gateway now: same cflx_ key, same OpenAI-compatible endpoint. You only change the model name.
Key takeaways
- GPT-6.1 Terra is OpenAI's new mid-tier model for agents that read screens and use a computer, at half the list price of GPT-5.6 Terra and GPT-6.1 Sol.
- Claude Opus 5.5 is Anthropic's recommended default for most work, at $4 / $20 per 1M tokens, below Opus 5.
- Claude Haiku 5.5 is the new cheap, fast model for classification, extraction and routing.
- You pay the provider's price plus a small margin, from one prepaid balance. GPT-5.6, GPT-6 Astra and Claude Opus 5 stay available.
The new models
| Model | Model name to send | Best for |
|---|---|---|
| GPT-6.1 Terra | gpt-6.1-terra | Agents that look at screens and act on a computer; strong visual reasoning at a mid-tier price |
| GPT-6.1 Sol | gpt-6.1-sol | OpenAI's main general model in the 6.1 line |
| GPT-6 Luna | gpt-6-luna | High-volume, low-cost calls |
| Claude Opus 5.5 | claude-opus-5-5 | Long-running agentic coding and knowledge work; Anthropic's suggested starting point |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | The balance of speed and quality |
| Claude Haiku 5.5 | claude-haiku-5-5 | Fast classification, extraction and routing |
All six read images and use tools, with context windows of about 1M tokens.
GPT-6.1 Terra in a sentence
OpenAI built Terra for agents that need to understand what's on a screen and get things done on a computer. OpenAI says it scores about 15 percentage points higher than GPT-5.6 Terra on Agents' Last Exam at roughly a third of the cost (medium reasoning), and its list price is $1 input / $5 output per 1M tokens, half of GPT-5.6 Terra and GPT-6.1 Sol.
What they cost through withConflux
Your price is the provider's price plus a small margin. OpenAI charges about twice as much when a request's input is over 272K tokens, and Haiku 5.5 costs more when a prompt is over 100K tokens. We meter those models at the higher rate, so a large request never runs anyone underwater. Claude Opus 5.5 and Sonnet 5.5 have one flat price at any length.
| Model | Provider list (per 1M, in / out) | Your price (per 1M, in / out) |
|---|---|---|
| GPT-6.1 Terra | $1 / $5 · over 272K: $2 / $7.50 | $2.50 / $9.38 |
| GPT-6.1 Sol | $2 / $10 · over 272K: $4 / $15 | $5.00 / $18.75 |
| GPT-6 Luna | $0.10 / $0.50 · over 272K: $0.20 / $0.75 | $0.25 / $0.94 |
| Claude Opus 5.5 | $4 / $20 | $5.00 / $25.00 |
| Claude Sonnet 5.5 | $2 / $10 | $2.50 / $12.50 |
| Claude Haiku 5.5 | $0.10 / $0.50 · over 100K: $0.50 / $2.50 | $0.63 / $3.13 |
Every model's live price is on the models page. It reads from the same catalog the gateway bills against, so what you see is what you're charged.
Which one to pick
Our standing advice hasn't changed: send hard, multi-step work to a strong model, and keep high-volume paths on a cheap one. With this release that means:
- Hard reasoning and long agent runs: Claude Opus 5.5 or GPT-6 Astra.
- Agents that use a screen or browser: GPT-6.1 Terra.
- Everyday product features: Claude Sonnet 5.5 or GPT-6.1 Sol.
- Classification, extraction, routing at volume: Claude Haiku 5.5 or GPT-6 Luna.
You don't commit to one. You pick per request, so trying Terra or Opus 5.5 on one path is a one-word change.
Managed or your own keys
On managed keys the new models work straight away, billed from your prepaid balance with your spend caps. On BYOK the call runs on your own OpenAI or Anthropic account, so the model has to be available on that account; the gateway still meters, dedupes and caps.
One key, one bill, every model, including the newest ones. That's the whole idea of the gateway.