LumeAPIShort path: Cheap LLM API · Multi-model API · OpenAI-compatible API · Models · AI API pricing · Models
Last verified: July 26, 2026
Open Cline settings, select OpenAI Compatible, set Base URL to https://api.lumeapi.site/v1, paste a LumeAPI API key, and enter an exact catalog model id such as gpt-5.6-terra. Cline officially documents Base URL, API Key, and Model ID as the three required fields for a generic OpenAI-compatible provider. Start with a low-risk repository task, verify the connection, and inspect the usage log before enabling long autonomous runs.
LumeAPI is an independent third-party gateway and is not affiliated with Cline, OpenAI, Anthropic, or Google.
Quick answer
| Setting | Value |
|---|---|
| API Provider | OpenAI Compatible |
| Base URL | https://api.lumeapi.site/v1 |
| API Key | A LumeAPI model API key, stored securely |
| Model ID | Exact id from /models, for example gpt-5.6-terra |
| First validation | One file edit, one command, one test, then inspect usage |
In short
Open Cline settings, select OpenAI Compatible, set Base URL to https://api.lumeapi.site/v1, paste a LumeAPI API key, and enter an exact catalog model id such as gpt-5.6-terra. Cline officially documents Base URL, API Key, and Model ID as the three required fields for a generic OpenAI-compatible provider.
Configure Cline step by step
- Open Cline and select the gear icon.
- Choose OpenAI Compatible as the API Provider.
- Enter
https://api.lumeapi.site/v1as Base URL. - Paste the LumeAPI model API key. Do not use the Research publishing secret.
- Enter
gpt-5.6-terraor another exact catalog id. - Set conservative output/context values until the model’s actual behavior is verified.
- Click Verify, then run a disposable repository task.
Cline’s documentation also exposes model configuration fields such as context window, max output tokens, image support, and input/output price. Treat those fields as client-side metadata: they do not add capabilities that the selected gateway route lacks.
Which model should Plan and Act use?
| Phase or task | Candidate route | Why |
|---|---|---|
| Cheap exploration or mechanical plan | gpt-5.4-mini | Lowest rate in this set; useful for bounded tasks |
| Routine plan and implementation | gpt-5.6-terra | Balanced default |
| Difficult implementation or recovery | gpt-5.6-sol | Higher-cost escalation |
| Claude route experiment | claude-sonnet-4-6 | Same OpenAI-compatible gateway surface, separate model economics |
| Premium Claude escalation | claude-opus-4-8 | Use only when accepted-task improvement justifies cost |
If your Cline version supports distinct Plan and Act model selection, use a cheaper route for exploration and promote only the execution phase that needs additional capability. If it does not, keep one model and cap turns rather than inventing an unsupported two-model configuration.
What does one feature implementation cost?
Assume a feature consumes 100,000 input tokens and 15,000 output tokens across repository reads, planning, edits, and test loops:
| Model | LumeAPI rate per 1M input/output | Calculated feature cost |
|---|---|---|
| GPT-5.4 mini | $0.225 / $1.35 | $0.0428 |
| GPT-5.6 Terra | $0.75 / $4.50 | $0.1425 |
| GPT-5.6 Sol | $1.50 / $9.00 | $0.2850 |
| Claude Sonnet 4.6 | $1.50 / $7.50 | $0.2625 |
| Claude Opus 4.8 | $2.50 / $12.50 | $0.4375 |
This table is not an observed Cline benchmark. Browser steps, MCP tools, terminal output, failed commands, and approval loops can materially increase tokens. Measure cost per accepted feature, not only per request.
Why Cline can burn more tokens than expected
An IDE agent repeatedly sees tool schemas, file context, diffs, command output, and conversation history. A verbose test failure can be more expensive than the final code. Browser or MCP use adds more tool rounds. Human approval can reduce unsafe actions but does not automatically reduce token use if the agent replans after every rejection.
Set a budget at the task level:
max turns → max input/output budget → required human checkpoint → escalation modelLog the model, turn count, tokens, billed amount, retries, and whether the feature was accepted. LumeAPI’s usage logs guide explains the evidence you need.
A realistic small-team workflow
A team can standardize one LumeAPI key policy while allowing multiple approved models. Use mini for mechanical chores, Terra for normal tickets, and Sol only for high-risk or failed work. Keep the provider configuration in team documentation, but distribute API keys through a secret manager rather than workspace settings committed to Git.
The economic advantage is operational: one account and usage view can cover approved GPT and Claude routes. It is not proof that every model behaves identically in Cline.
Troubleshooting
Cline says Invalid API Key
Confirm that the key belongs to the LumeAPI model API, that it has not been copied with whitespace, and that the Base URL is correct.
Cline says Model Not Found
Use the exact current catalog id. Do not add Aider’s openai/ prefix in the Cline Model ID field unless the LumeAPI catalog itself contains that prefix.
Verify fails with a connection error
Confirm the URL ends in /v1, check network policy, and test the same key/model with the catalog’s curl example.
Tool calls fail even though chat works
A successful text response does not prove tool compatibility. Reduce the test to one simple tool and capture the request/response error without exposing secrets.
What most guides get wrong
The usual mistake on How Do You Configure Cline to Use LumeAPI as an OpenAI-Compatible Provider? is to chase the lowest posted per-million-token rate while ignoring orchestration: repeated context, oversized tool output, silent retries, and premium-model defaults often dominate the invoice.
A cheaper list price does not help when every agent hop resends the same repository map or when failed tests trigger unbounded re-runs. Optimize cost per successful task—not the price of a single completion—before you renegotiate vendor contracts or swap gateways.
A realistic production scenario
A six-person platform team standardized on one premium model for How Do You Configure Cline to Use LumeAPI as an OpenAI-Compatible Provider? 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 this guide apply to every LumeAPI model?
Verify the exact model id and rate on /models before production use for How Do You Configure Cline to Use LumeAPI as an OpenAI-Compatible Provider?.
Is Cline open source?
The cline/cline repository is public and used the Apache-2.0 license when checked July 26, 2026.
Can Cline use Claude through LumeAPI?
Cline can send a LumeAPI Claude catalog id through its OpenAI-compatible provider. Verify tool behavior for the exact route before production use.
Is Cline the same as Claude Code?
No. Cline is an open-source agent surface with configurable providers; Claude Code is Anthropic’s coding agent with its own gateway protocol.
Sources and testing boundary
- Cline OpenAI Compatible provider documentation
- Cline GitHub repository — 65,058 stars observed July 26, 2026
- LumeAPI model catalog
- Multi-model production checklist
No billable Cline task was run for this report. Before production use, capture the Cline version, Verify result, one successful tool-driven edit, and its matching LumeAPI usage record.