GPT-5.6 Luna Drops 80% to $0.20/M Input — OpenAI's Response to Claude Opus 5
OpenAI just dropped GPT-5.6 prices by up to 80%, one day after Anthropic launched Claude Opus 5 at $5/$25. Luna, their fastest model, now costs $0.20 per million input tokens and $1.20 per million output tokens. That is 80% cheaper than yesterday and 25x cheaper than Opus 5.
What changed
GPT-5.6 Luna (the speed tier):
- Old: $1.00/$6.00 per 1M tokens
- New: $0.20/$1.20 per 1M tokens
- Reduction: 80%
GPT-5.6 Terra (the balanced tier):
- Old: $2.50/$15.00 per 1M tokens
- New: $2.00/$12.00 per 1M tokens
- Reduction: 20%
GPT-5.6 Sol (the frontier tier):
- Price: $5.00/$30.00 (unchanged)
- New feature: Fast mode (2.5x speed at 2x price, replacing Priority Processing)
The new pricing landscape
Here is how the major models stack up after this cut:
| Model | Input/1M | Output/1M | Key Metric |
|---|---|---|---|
| GPT-5.6 Luna | $0.20 | $1.20 | 84.3% Terminal-Bench |
| DeepSeek V4 Flash | $0.14 | $0.28 | 79.0% SWE-bench |
| Qwen 3.6 Flash | $0.25 | $1.00 | Free tier available |
| Claude Sonnet 5 | $2.00 | $10.00 | 63.2% SWE-bench Pro |
| GPT-5.6 Terra | $2.00 | $12.00 | 82.5% Terminal-Bench |
| Claude Opus 5 | $5.00 | $25.00 | 43.3% Frontier-Bench |
| GPT-5.6 Sol | $5.00 | $30.00 | 88.8% Terminal-Bench |
| Claude Fable 5 | $10.00 | $50.00 | 33.7% Frontier-Bench |
Luna at $0.20/$1.20 with 84.3% Terminal-Bench is now the clear value king. It costs 25x less than Opus 5 on input and 20x less on output, while delivering competitive performance on most benchmarks.
Why OpenAI did this
Timing is not a coincidence. Anthropic launched Claude Opus 5 yesterday at $5/$25, which immediately made GPT-5.6 Sol ($5/$30) look overpriced. OpenAI needed to respond.
The Luna cut is the aggressive move. At $0.20/$1.20, OpenAI is clearly positioning Luna as the default model for high-volume workloads. They want developers to use Luna for everything that does not need frontier intelligence, and only escalate to Sol when necessary.
This is the same strategy Anthropic uses with Sonnet 5 ($2/$10) vs Opus 5 ($5/$25), but OpenAI is pushing the price floor much lower.
What this means for developers
If you are using Claude Opus 5: Keep using it for complex tasks. Opus 5 still leads on Frontier-Bench (43.3% vs Sol’s comparable range) and has the 5-level effort control system. But for routine work, consider switching to Luna at $0.20/$1.20.
If you are using Claude Sonnet 5: Luna at $0.20/$1.20 is now 10x cheaper on input and 8x cheaper on output. Luna scores higher on Terminal-Bench (84.3% vs Sonnet 5’s competitive range). Unless you need Anthropic-specific features, Luna is the better value.
If you are using DeepSeek V4 Flash: DeepSeek is still cheaper ($0.14/$0.28 vs $0.20/$1.20) but Luna scores higher on most benchmarks. The gap between them has narrowed significantly.
If you are building high-volume APIs: Luna at $0.20/$1.20 makes AI-powered features economically viable in scenarios that were previously too expensive. Customer support, content moderation, data classification, and similar workloads can now run at scale for pennies.
The Fast mode addition
OpenAI also introduced Fast mode for GPT-5.6 Sol, replacing Priority Processing. Fast mode delivers 2.5x faster speeds at 2x the price ($10/$60 per 1M tokens). This is the same pricing model as Anthropic’s fast mode for Opus 5.
Existing Priority Processing API requests will automatically use Fast mode. No code changes needed.
Updated pricing articles
We have updated the following articles with the new pricing:
What to watch next
- Anthropic response. Will they cut Sonnet 5 or Opus 5 prices? The pressure is real.
- Google response. Gemini 3.6 Flash at $1.50/$7.50 is now significantly more expensive than Luna.
- DeepSeek response. Their $0.14/$0.28 pricing advantage over Luna is now marginal.
- Community reaction. Luna at $0.20/$1.20 with 84.3% Terminal-Bench will change how developers think about model selection.
For the full pricing breakdown, see our AI API pricing comparison. For model details, see the GPT-5.6 complete guide.