LumeAPIShort path: Cheap LLM API · Cheap Claude API · Claude API · AI API pricing
Last verified: July 26, 2026
Not as a documented direct connection today. Claude Code supports ANTHROPIC_BASE_URL, but Anthropic requires the gateway to expose a supported Anthropic-format endpoint and forward Claude Code’s protocol correctly. LumeAPI’s public catalog currently documents its Claude model ids on the OpenAI-compatible chat/completions surface. An OpenAI-compatible base URL alone is therefore not evidence that Claude Code can call it directly. Use a verified translation gateway that presents Anthropic format to Claude Code, or wait for LumeAPI to document native Claude Code gateway compatibility.
LumeAPI is an independent third-party gateway and is not affiliated with Anthropic. This page intentionally answers the protocol question before discussing price.
Quick answer
| Question | Verified answer |
|---|---|
Can I paste https://api.lumeapi.site/v1 into ANTHROPIC_BASE_URL? | Do not assume so; the published LumeAPI contract is OpenAI Chat Completions |
| What does Claude Code expect from a gateway? | A supported API format, including an Anthropic-format endpoint where required |
| Can a translation gateway sit between them? | Architecturally yes, but the exact bridge must be tested end to end |
| Which LumeAPI Claude ids are currently listed? | claude-sonnet-4-6 and claude-opus-4-8 among the current catalog routes |
| Are the session costs below benchmark results? | No; they are transparent token scenarios only |
In short
Not as a documented direct connection today. Treat published catalog rates as a planning input and confirm them on /ai-api-pricing before changing production traffic.
Why protocol compatibility matters
Claude Code is not a generic chat box. It sends agent turns, tool definitions, streaming events, usage fields, and evolving Claude Code-specific headers through its supported gateway protocol. Anthropic warns that a gateway which fails to forward new capabilities can break corresponding Claude Code features.
LumeAPI currently lists Claude Sonnet 4.6 and Claude Opus 4.8 with chat/completions as the endpoint. That is useful for OpenAI SDK-compatible applications, but it does not by itself satisfy Claude Code’s Anthropic-format gateway contract.
The safe architecture
Use this topology only after staging verification:
Claude Code
│ Anthropic-format requests
▼
Protocol-aware gateway or translator
│ OpenAI-compatible Chat Completions
▼
https://api.lumeapi.site/v1
│
├─ claude-sonnet-4-6
└─ claude-opus-4-8The bridge must translate messages, tool calls, streaming events, errors, token usage, and model names. A simple reverse proxy that only changes the host is insufficient.
Anthropic documents ANTHROPIC_BASE_URL for a compatible gateway and recommends per-user gateway credentials. A future verified configuration would resemble this on the Claude Code side:
export ANTHROPIC_BASE_URL=https://your-anthropic-format-bridge.example
export ANTHROPIC_AUTH_TOKEN=your-gateway-token
claude --model your-bridge-model-aliasThis is a bridge-side example, not a direct LumeAPI configuration. Do not set it to the LumeAPI /v1 URL until an Anthropic-format endpoint is documented and tested.
Which Claude route would a translated workflow use?
| Task | Candidate LumeAPI route | July 26 LumeAPI rate |
|---|---|---|
| Routine repository work | claude-sonnet-4-6 | $1.50 / $7.50 per 1M input/output tokens |
| Expensive escalation | claude-opus-4-8 | $2.50 / $12.50 per 1M input/output tokens |
The model id seen by Claude Code may be an alias owned by the translation gateway. The upstream request must ultimately use an exact LumeAPI catalog id.
What would a terminal-agent session cost?
| Scenario | Assumed tokens | Sonnet 4.6 | Opus 4.8 |
|---|---|---|---|
| Short 30-minute task | 60k input / 8k output | $0.150 | $0.250 |
| Medium 60-minute task | 120k input / 16k output | $0.300 | $0.500 |
| Long 120-minute task | 240k input / 32k output | $0.600 | $1.000 |
These figures exclude bridge fees, retries, prompt caching differences, tool-result amplification, and failed runs. Duration is only a label for the token assumption; Claude Code is billed by usage, not by minutes.
What most guides get wrong
The most damaging mistake is treating ANTHROPIC_BASE_URL as a universal “custom OpenAI endpoint” switch. It changes the destination, not the protocol Claude Code speaks. A second mistake is pricing only the final answer while ignoring file reads, terminal output, tool schemas, and repeated context.
A third mistake is mapping “Sonnet” or “Opus” to a gateway id without testing tool calling. A successful one-line chat completion does not prove that an agentic edit loop works.
Production acceptance test
Before rollout, capture all of the following:
- Claude Code version and bridge version.
- A file-read, file-edit, shell-command, and test-run sequence.
- Streaming behavior and tool-call round trips.
- Model id received by LumeAPI.
- Input/output usage in the usage log.
- Behavior on 401, 429, timeout, and tool error.
- Cost per accepted task, including failed attempts.
For cost background, see Claude API cost reduction and Sonnet versus Opus pricing.
A realistic production scenario
A six-person platform team standardized on one premium model for Can Claude Code Connect Directly to LumeAPI’s OpenAI-Compatible API? after a strong demo score. Early invoices looked fine; by week three, failing test loops had doubled input tokens without more accepted output.
They kept the orchestrator but tightened routing: a lower-cost catalog model handled exploration, a mid-tier model handled implementation, and premium routes required a logged failure condition. They truncated test logs and stopped attaching full diffs on every retry.
The breakthrough was reconciling LumeAPI usage records with task ids. Token list price mattered less than eliminating duplicate context across retries.
Expert take
LumeAPI is an independent third-party OpenAI-compatible gateway—not OpenAI, Anthropic, or Google. Published catalog rates can be materially below official reference pricing, but compatibility does not guarantee identical behavior for every native feature, streaming edge case, tool schema, Batch workflow, or prompt-cache semantics.
Follow this advice when you control prompts, retries, and routing and need auditable USD billing across GPT, Claude, and Gemini on one key. Do not cut over blindly if you depend on provider-native capabilities your stack has not tested end to end.
Run a bounded smoke test on your exact SDK, model id, streaming mode, and tool path; reconcile usage logs with orchestrator task ids; and keep a fallback route until error rates and latency are acceptable.
FAQ
Does Claude Code officially support third-party gateways?
Yes, when the gateway exposes a supported API format. Anthropic does not endorse or maintain third-party gateway products.
Does OpenAI compatibility automatically satisfy Claude Code?
No. OpenAI Chat Completions and Anthropic Messages are different protocol contracts.
Can Claude Code route non-Claude models through a gateway?
Anthropic’s current documentation says it does not support routing Claude Code to non-Claude models through a gateway.
Can I still use LumeAPI Claude routes from Python or Node.js?
Yes, through the documented OpenAI-compatible route. See using Claude with an OpenAI SDK.
Sources and testing boundary
- Anthropic: Other LLM gateways
- Anthropic Claude Code repository — 139,116 stars observed July 26, 2026
- LumeAPI model catalog
No Claude Code bridge or billable LumeAPI agent session was executed for this report. Accordingly, this page does not claim that LumeAPI is a direct Claude Code endpoint; teams should run the acceptance test above before deploying an Anthropic-format translation gateway.