OpenAI Cuts GPT-5.6 Luna Prices 80% and Terra 20% After Sol Optimises Its Own Inference

OpenAI dropped prices across two tiers of its GPT-5.6 family on July 30, with Luna falling 80% and Terra falling 20%, effective immediately across the API, ChatGPT, Codex, and partner platforms including Amazon Bedrock and Netlify AI Gateway.

The reductions came from efficiency gains in OpenAI's inference stack — specifically GPU kernel optimisations and an expanded use of speculative decoding, the same technique that lets a smaller draft model propose tokens that the full model then verifies in fewer passes. OpenAI attributed a meaningful share of those gains to Sol, its flagship tier, which the company said helped identify and refine its own inference bottlenecks.

New pricing

| Model | Input (per 1M tokens) | Output (per 1M tokens) | Change | |---|---|---|---| | GPT-5.6 Luna | $0.20 | $1.20 | -80% | | GPT-5.6 Terra | $2.00 | $12.00 | -20% | | GPT-5.6 Sol (standard) | $5.00 | $30.00 | Unchanged | | GPT-5.6 Sol (Fast mode) | $10.00 | $60.00 | New tier |

The new Sol Fast mode replaces the previous Priority Processing tier and delivers up to 2.5× higher throughput — useful for latency-sensitive production workloads that can tolerate the higher per-token cost.

Luna's new price point puts it among the most aggressive in its category: at $0.20 input, it undercuts most mid-tier competitor offerings while retaining the GPT-5.6 architecture. The cut is likely to increase pressure on competitors at the fast, high-volume end of the market — the segment where Gemini 3.5 Flash-Lite and DeepSeek-V4-Flash are already competing hard on cost.

Sol pricing unchanged reflects OpenAI's read that frontier-tier demand is relatively price-inelastic at this stage — the enterprises running complex agentic tasks on Sol are not the same customers who were waiting for a Luna price drop to scale.