GPT-6 Sol and Luna launch: API prices permanently cut 50%, which model should developers pick?

TL;DR

If you searched for "GPT-6 Sol Luna API pricing" or "how to migrate from GPT-5.6 to GPT-6": both models launched September 22, 2026, with API prices permanently cut 50% vs GPT-5.6 promotional rates. Sol handles complex coding and agent tasks; Luna is built for high-frequency, low-cost workloads.

Pricing Comparison (per million tokens, USD)

Note: GPT-5.6 Sol's $4/$20 was already a promotional rate guaranteed only through November 21, 2026. The real saving vs list price is greater than 50%.

Model Positioning

Migration Breaking Changes

1. Function calling behavior change

In Chat Completions, function calling only works with reasoning_effort: "none". For any other effort level, migrate tool-calling code to the Responses API.

2. Sampling parameters removed

When reasoning is on, drop temperature, top_p, top_logprobs, and logprobs.

3. Reasoning effort renaming

If you used minimal effort on older models, start with low on GPT-6 and benchmark.

4. Prompt caching config changed

Old: prompt_cache_retention: {"ttl": "30m"}

New: prompt_cache_options: {"ttl": "30m"}

Caching starts at 1,024 visible input tokens. Cache writes cost 125% of input price; reads cost 10%.

5. Codex auto-switching

Multiple developers report Codex sessions being automatically switched to GPT-6 models. Pin the model ID explicitly if you depend on GPT-5.6 behavior.

Developer Model Selection Guide

Reminder: OpenAI has not announced GPT-5.6 API deprecation dates. But Codex is already auto-switching, so evaluate migration early.

Next Steps

1. Replace existing model IDs with gpt-6-sol and gpt-6-luna in your test environment
2. Check if your function-calling code depends on Chat Completions (if so, migrate to Responses API)
3. Use caching to optimize costs: a stable system prompt saves 90% on input from the second call
4. Track real-time pricing at OpenAI API Pricing

Related tutorials: how to build an automated content publishing pipeline with Claude Code and n8n at /ai-agent-tutorials.