Anthropic & OpenAI Compatible

Use PublicaAI from your code

PublicaAI exposes both Anthropic-compatible (/v1/messages) and OpenAI-compatible (/v1/chat/completions) endpoints backed by Claude on AWS Bedrock. Works with Anthropic SDK, OpenAI SDK, LangChain, Claude Code, Cursor, and more.

Base URL

https://bedrock.publicaai.com

Auth header

Authorization: Bearer sk-…

Available models

Model ID	Description	Thinking	Cost multiplier	Max tokens
`claude-5-5-opus`	Latest Opus — most capable reasoning and agentic model	adaptive	5×	500,000
`claude-5-5-sonnet`	Latest Sonnet — fast, near-Opus quality	adaptive	2×	500,000
`claude-5-5-haiku`	Latest Haiku — fastest and most affordable	—	1×	500,000
`gpt-6.1-sol`	GPT 6.1 Sol — latest OpenAI flagship	—	5×	32,000
`gpt-6-sol`	GPT 6 Sol	—	4×	32,000
`gpt-6-luna`	GPT 6 Luna	—	4×	32,000
`gpt-6-astra`	GPT 6 Astra	—	4×	32,000
`gpt-5.6-sol`	GPT 5.6 Sol	—	3×	32,000
`gpt-5.6-luna`	GPT 5.6 Luna	—	3×	32,000
`gpt-5.6-terra`	GPT 5.6 Terra	—	3×	32,000
`gpt-5.5`	GPT 5.5	—	2.5×	32,000
`gpt-5.4`	GPT 5.4	—	2×	32,000
`claude-5-opus`	Flagship — deep reasoning and complex agentic coding	adaptive	5×	128,000
`claude-5-sonnet`	Best speed-to-intelligence ratio; near-Opus quality	adaptive	2×	128,000
`claude-4-8-opus`	Most capable Opus 4 model — agentic and knowledge work	adaptive	5×	128,000
`claude-4-7-opus`	Highly autonomous, strong long-horizon agentic work	adaptive	5×	128,000
`claude-4-6-opus`	Most capable, highest quality	adaptive + budget	5×	300,000
`claude-4-6-sonnet`	Enhanced performance and capabilities	adaptive + budget	3.5×	300,000
`claude-4-5-opus`	Previous-generation Opus	budget	5×	64,000
`claude-4-5-sonnet`	Balanced performance	budget	3×	4,000
`claude-4-5-haiku`	Fast, low cost	—	1×	4,000

Credit cost: (input_tokens + output_tokens × 5) × cost_multiplier × 1.2. Output tokens are 5× more expensive than input tokens.

adaptive models (Opus 4.7 and newer, Sonnet 5) decide how much to reason on their own — send thinking: { type: "adaptive" }. They no longer accept a temperature or a budget_tokens value; the gateway strips both and upgrades legacy budget_tokens requests to adaptive automatically, so existing client code keeps working.

Quickstart

Install the Anthropic SDK and point it at PublicaAI:

bash

pip install anthropic

python

from anthropic import Anthropic

client = Anthropic(
    api_key="sk-your-publicaai-key",     # your staff API key
    base_url="https://bedrock.publicaai.com",
)

message = client.messages.create(
    model="claude-5-sonnet",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Summarise our company values in 3 bullets."},
    ],
)

print(message.content[0].text)
print(f"Credits remaining: {message.usage.credits_remaining}")

API Endpoints

POST

/v1/messages

Recommended

Anthropic-compatible Messages API. Use with Anthropic SDK or any tool expecting native Anthropic format (Claude Code, Cursor, etc.).

POST

/v1/chat/completions

OpenAI-compatible Chat Completions API. Use with OpenAI SDK, LangChain, or any OpenAI-compatible tool.

GET

/v1/models

Lists every model available to your key, with cost multipliers and limits.

GET

/v1/usage

Returns your remaining credits, 24h usage breakdown and per-model statistics.

Limits & responses

Rate limit

requests / minute / key

Max tokens

8,000

per request (Opus/Sonnet 4.6)

Insufficient credits

402

HTTP status returned