OpenAI's Luna AI Just Got 80% Cheaper on Amazon Bedrock
The cheapest AI isn't always the cheapest answer — here's why.
Forget dollars per million tokens. Production workloads don't buy tokens — they buy outcomes, and the pricing page ignores the multipliers that decide the bill: how often a model is right, how many tokens it needs to get there, and, for agentic workloads, how many turns it takes. That's the case made in a post sharing results from an open-source benchmarking harness that ran one identical code path (the OpenAI Responses API) against OpenAI models on Amazon Bedrock (gpt-5.6-luna, gpt-5.6-terra, gpt-5.6-sol) and two cost-efficient OpenAI API models (gpt-5.4-mini, gpt-5.4-nano) used as cost-optimized baselines. At its original list price (roughly 1.5× mini's), luna was already 25 percent cheaper per correct AIME answer in this configuration, because with reasoning disabled it used fewer billed tokens than mini at its defaults. Then came the July 30, 2026 price reduction for GPT-5.6 Luna and Terra on Amazon Bedrock — luna down 80 percent, terra down 20 percent. Recorded cost per correct AIME answer: $0.0021 for luna versus $0.0139 for mini. One caveat from the authors: this compares practical deployment configurations, not intrinsic model capability — reproduce it on your own workload before choosing.
- Cheaper AI models can cost more overall, because you pay for wrong answers and retries too.
- OpenAI's Luna dropped 80% in price on Amazon's AI marketplace on July 30, 2026; Terra fell 20%.
- The top model, Sol, solved 75% of hard math problems versus Mini's 37% — a real gap in usefulness.
Why It Matters
Businesses overpaying for the wrong AI model pass those costs to you — in prices and slower service.