OpenAI cuts GPT-5.6 Luna 80% to $0.20/$1.20 as users hit 1B
Terra falls 20% to $2/$12, Sol gains a 2.5x Fast mode, and the savings flow into Codex credits — while Grok Voice auto-migrates its API default tomorrow.
Copy markdown
Luna is now 80% cheaper: $0.20 in, $1.20 out
GPT-5.6 Luna dropped from $1/$6 to $0.20/$1.20 per million tokens. A high-throughput agent or RAG pipeline that was burning ~$600/day on Luna now runs closer to $120 — cheap enough to make it your default for bulk classification, extraction and tool-call routing.
Terra down 20%, and Sol gets a 2.5x Fast mode
GPT-5.6 Terra fell to $2/$12 (from $2.50/$15), and Sol added a Fast mode that runs up to 2.5x standard speed at double the price. That's a per-request latency lever — pay up only when a reasoning-heavy call is blocking a user.
The cuts land inside Codex and ChatGPT Work
Cheaper Luna and Terra tokens stretch the same credit budget under Codex and ChatGPT Work subscriptions — quotas and plan prices didn't change, so your effective rate limit just went up for free. Worth re-checking any hard-coded cost estimates in your billing dashboards.
Why now: 1B users, and OpenAI's own stack got cheaper
OpenAI disclosed over 1 billion active users and 2M+ businesses, crediting the cuts to efficiency — an autonomous GPU-kernel optimizer saving ~20% and speculative-decoding gains above 15%. Read it as a signal the price floor keeps falling, so avoid locking in long commitments at today's rates.
Heads up: Grok Voice flips its default tomorrow
On Aug 5, grok-voice-latest auto-migrates to Think Fast 2.0 ($0.08/min): time-to-first-audio drops to 0.70s from 1.25s and transcription runs 1.5-2x better than Deepgram Nova 3 and ElevenLabs Scribe v2. If your voice agent depends on 1.0's timing, pin grok-voice-think-fast-1.0 before then.