Claudexia serves nine models through one endpoint and one key. Here's the full list with exact prices — no rounding, no "starting at."
The full price list
| Model | Input, $/1M | Output, $/1M |
|---|---|---|
claude-opus-5 | 0.4 | 0.4 |
claude-opus-4.8 | 0.4 | 0.4 |
claude-opus-4.7 | 0.4 | 0.4 |
claude-opus-4.6 | 0.4 | 0.4 |
claude-sonnet-5 | 0.3 | 0.3 |
claude-sonnet-4.6 | 0.3 | 0.3 |
gpt-5.6-sol | 0.3 | 0.3 |
gpt-5.6-terra | 0.2 | 0.2 |
gpt-5.6-luna | 0.1 | 0.1 |
There's no Haiku tier in the lineup. The cheapest model is gpt-5.6-luna, and it's a separate family, not Claude. The request format is compatible; behavior and prompts are worth checking against your own task.
The flagship: Claude Opus 5
claude-opus-5 shipped on July 24, 2026, and is the current flagship. Reach for it on tasks that can't be broken into clear steps ahead of time: reading an unfamiliar architecture, debugging without a working hypothesis, planning under conflicting constraints.
The honest part about four Opus revisions
claude-opus-5, claude-opus-4.8, claude-opus-4.7, and claude-opus-4.6 are all priced the same: $0.4 in, $0.4 out, each of them. We don't mark up older revisions and we don't discount newer ones. If your system runs on 4.7 and it works, there's no money on the table by moving off it. Choosing between Opus revisions is a behavioral call for your task, not a pricing one.
Same story with Sonnet: claude-sonnet-5 and claude-sonnet-4.6 are both $0.3. A newer revision doesn't mean paying more, an older one doesn't mean saving.
Decision table by task
| Task | Model | Why |
|---|---|---|
| Bulk classification, log parsing | gpt-5.6-luna | Cheap, the task doesn't need multi-step reasoning |
| Data extraction, format conversion | gpt-5.6-luna or gpt-5.6-terra | Move to Terra if Luna misses too often |
| Summarization, Q&A over mid-size documents | claude-sonnet-5 or gpt-5.6-sol | Same price, pick by output style |
| Agentic coding, refactors that follow a plan | claude-sonnet-5 | Holds long context, follows instructions precisely |
| Stable production with calibrated prompts | claude-sonnet-4.6 or claude-opus-4.6 | Behavior proven over months of traffic |
| Unfamiliar architecture, hard debugging, deep planning | claude-opus-5 | Flagship, where reasoning depth matters most |
One integration, any model
from anthropic import Anthropic
client = Anthropic(
api_key="sk_cdx_your_key_here",
base_url="https://api.claudexia.tech",
)
for model in ["gpt-5.6-luna", "claude-sonnet-5", "claude-opus-5"]:
message = client.messages.create(
model=model,
max_tokens=256,
messages=[{"role": "user", "content": "Compress this text into one paragraph."}],
)
print(model, "->", message.content[0].text[:80])
The same loop works over OpenAI Chat Completions at https://api.claudexia.tech/v1 — only the request body shape changes, the key and model names stay the same.
Two formats, one key
Both Anthropic Messages and OpenAI Chat Completions run on the same Claudexia endpoint. No need to keep a second key for a different SDK or split infrastructure across two providers just to match a request format.
Budgeting ahead of time
The site has a calculator: pick a model, enter your input and output volume, see the total. It's worth running the same scenario across a few models and seeing the difference in real numbers instead of a gut feeling. Stats run per key, and limits and thresholds are configurable per user and per team, so overspend shows up before it's a problem.
In short
Nine models, three price tiers: $0.1 for Luna, $0.2 for Terra, $0.3 for Sonnet and Sol, $0.4 for all four Opus revisions. Within each tier the price is identical, so the choice always comes down to how a model behaves on your task, not what it costs.