OpenAI Slashes Prices for GPT-5.6

⚡

Key Takeaways

1OpenAI has announced a 20% reduction for GPT-5.6 Terra and an 80% drop for GPT-5.6 Luna.

2GPT-5.6 Sol has optimized inference and forward pass, reducing service costs by 20%.

3Luna is now more competitive than Google's Gemini 3.1 Flash-Lite, with significantly lower rates.

💡Why it matters — These price cuts reposition OpenAI as a leader in the affordable AI model market, directly challenging Google and Anthropic.

OpenAI Announces Massive Price Cuts for GPT-5.6

On July 30, 2026, OpenAI made headlines by unveiling a significant price reduction for its GPT-5.6 models. The GPT-5.6 Terra model sees a 20% decrease, while GPT-5.6 Luna experiences a dramatic 80% drop in cost.

Optimization through GPT-5.6 Sol

This major advancement is attributed to the introduction of GPT-5.6 Sol, which has significantly improved the model's efficiency. OpenAI explained in a detailed article how GPT-5.6 Sol was used to optimize the forward pass, a crucial process that transforms inputs into predictions of subsequent tokens. While individual operations are fast, inefficiencies such as excessive memory movement and improper synchronization can slow down GPUs. GPT-5.6 Sol has enabled the identification and correction of these inefficiencies by pre-computing, avoiding, or parallelizing certain tasks. Thanks to Codex, GPT-5.6 Sol also rewrote and optimized the production kernels, the core code that executes the model's mathematical operations. This was made possible because GPT-5.6 was trained to excel in writing and improving kernels in Triton and Gluon, two open-source GPU programming languages supported by OpenAI. These efforts have led to a 20% reduction in end-to-end service costs.

A New Landscape for Low-Cost Models

The price cut for Luna redefines the market for low-cost AI models. With a rate of $0.20 per million tokens for input and $1.20 for output, Luna becomes more affordable than Google's Gemini 3.1 Flash-Lite model, which costs $0.025 for input and $1.50 for output.

Comparison with Competitors

In comparison, Anthropic's cheapest model, Claude Haiku 4.5, is priced at $1 for input and $5 for output. Luna is now five times cheaper for input compared to Claude Haiku 4.5, whereas previously, the two models had similar pricing.

Rapid Adoption by Users

In light of these new pricing conditions, many users are turning to Luna. For instance, a demo site, agent.datasette.io, which previously used Gemini 3.1 Flash-Lite, has already migrated to Luna to take advantage of these substantial savings.