LumeAPILast verified: August 15, 2026
Short path: AI API pricing · Cheap LLM pick list · GPT-5.6 Terra · Models
LumeAPI is an independent third-party gateway. This page owns a normalized token workload across GPT, Claude, and Gemini. Live catalog UI: /ai-api-pricing. Budget picks (which id to start on): /cheap-llm-api. OpenRouter's 5.5% credit fee is not this page — see the OpenRouter cost article.
Quick Answer
Normalized shape used below: 1.0 million input + 0.2 million output tokens, standard real-time text, no cache, no Batch, no tools.
| Model | Official $/1M in/out (this date) | LumeAPI $/1M in/out | Cost on the 1.0 + 0.2 shape (LumeAPI) |
|---|---|---|---|
| GPT-5.4 mini | $0.75 / $4.50 | $0.225 / $1.35 | $0.495 |
| Gemini 3 Flash | $0.50 / $3.00 | $0.25 / $1.50 | $0.550 |
| GPT-5.6 Terra | $2.00 / $12.00 live OpenAI | $0.75 / $4.50 | $1.65 |
| Gemini 3.5 Flash | $1.50 / $9.00 | $0.75 / $4.50 | $1.65 |
| Claude Sonnet 4.6 | $3.00 / $15.00 | $1.50 / $7.50 | $3.00 |
| GPT-5.6 Sol | $5.00 / $30.00 | $1.50 / $9.00 | $3.30 |
| Claude Opus 5 | $5.00 / $25.00 | $2.50 / $12.50 | $5.00 |
- Official rates from provider pages dated August 15, 2026. LumeAPI rates from catalog sync 2026-07-22.
- Quality is not in the table. A $0.495 row that fails eval is more expensive than a $1.65 row that passes once.
- Models listed by the provider but absent from LumeAPI (GPT-5.6 Luna, Claude Sonnet 5, Gemini 3.6 Flash) have no LumeAPI price on this page.
In short
Same token shape, two columns, dated sources. LumeAPI is cheaper on listed ids at standard rates. It is not cheaper on Batch, cache, Grounding, or models it does not sell.
What most guides get wrong
They apply one “70% off” badge to every row. GPT catalog badges are 70% off LumeAPI's stored official_price string. OpenAI's live Terra rate is now $2 / $12, while catalog.py still stores Terra official_price as $2.50 / $15.00. Using the live official rate, Terra on LumeAPI ($0.75 / $4.50) is 62.5% below OpenAI, not 70%.
This page shows both numbers. The commercial table on /ai-api-pricing follows the catalog string until that file is updated.
Terra official-price discrepancy (do not hide it)
| Source | GPT-5.6 Terra in / out per 1M |
|---|---|
| OpenAI API pricing (verified 2026-08-15) | $2.00 / $12.00 |
LumeAPI official_price field (catalog sync 2026-07-22) | $2.50 / $15.00 |
LumeAPI billed catalog price | $0.75 / $4.50 |
OpenAI cut Terra 20% on July 30, 2026 (OpenAI announcement). LumeAPI's billed Terra rate did not change in the 2026-07-22 catalog snapshot. Recheck both catalogs before a purchase order.
How the 1.0 + 0.2 column is calculated
cost = 1.0 × input_$ + 0.2 × output_$Example, LumeAPI Terra: 1.0 × 0.75 + 0.2 × 4.50 = $1.65.
Official OpenAI Terra on the same shape: 1.0 × 2.00 + 0.2 × 12.00 = $4.40.
Official Anthropic Sonnet 4.6: 1.0 × 3.00 + 0.2 × 15.00 = $6.00 vs LumeAPI $3.00.
Models this page will not price
Do not invent a LumeAPI rate for:
| Provider lists it | Official note (2026-08-15) | In LumeAPI text catalog? |
|---|---|---|
| GPT-5.6 Luna | OpenAI $0.20 / $1.20 short-context | No |
| Claude Sonnet 5 | Anthropic $2 / $10 standard | No |
| Gemini 3.6 Flash | Google $1.50 / $7.50 standard | No |
Use the provider directly, or pick a listed id from /models.
A realistic production scenario
100 document tasks on LumeAPI Terra: 4.0M input + 0.6M output tokens.
4.0 × $0.75 + 0.6 × $4.50 = $5.70If 92 tasks pass a pre-declared eval: $5.70 / 92 = $0.0620 per accepted task. A mini row at $0.495 on the 1.0+0.2 shape is irrelevant until you run this eval set. That is the only comparison that should change routing.
Smoke-test the id you put in the spreadsheet:
curl https://api.lumeapi.site/v1/chat/completions \
-H "Authorization: Bearer $LUMEAPI_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"gpt-5.6-terra","messages":[{"role":"user","content":"ping"}],"max_tokens":16}'Expert take
Publish the workload shape, the two rate sources, and the date. Then run an eval. The cheapest row in the table is a starting candidate, not a production default. When OpenAI, Anthropic, or Google change a list price, recompute the 1.0+0.2 column — do not reuse last month's percentage badge.
FAQ
What is the best AI API pricing comparison for 2026?
This Research page is the dated cross-provider worksheet. The live Owner table is /ai-api-pricing. Budget starting ids: /cheap-llm-api.
Which listed model is cheapest on the 1.0 + 0.2 shape?
gpt-5.4-mini at $0.495 on LumeAPI. It is the production choice only if it passes your eval.
Why is Terra 62.5% off here and 70% on the catalog badge?
The badge uses catalog official_price $2.50 / $15. This page uses OpenAI's live $2 / $12. Both are labeled.
Is Claude Sonnet 5 in the table?
No. It is $2 / $10 on Anthropic and not in the LumeAPI catalog. No invented LumeAPI cell.
Do Batch or cache change the ranking?
Yes. Recalculate in that billing mode. Do not apply Batch 50% to synchronous Chat Completions.
Sources and methodology
- OpenAI API pricing — Sol $5 / $30, Terra $2 / $12, Luna $0.20 / $1.20 short-context; verified August 15, 2026.
- Anthropic platform pricing — Opus 5 $5 / $25, Sonnet 4.6 $3 / $15, Sonnet 5 $2 / $10.
- Google Gemini API pricing — 3.5 Flash $1.50 / $9, 3 Flash $0.50 / $3, 3.1 Pro $2 / $12 (≤200k).
- LumeAPI <code>/api/models</code> and pricing hub — billed rates; catalog sync 2026-07-22.
Normalized costs are 1×input + 0.2×output from those rows. No cache, Batch, tax, or negotiated discount.