July 9, 2026

GPT-5.6 Is Live, and Luna Costs 25% Less in the Docs Than on the Launch Page

The family is live in the API, ChatGPT, and Codex, but OpenAI's own pages disagree on the cheapest tier's price and context window. Plus GPT-Live voice shipped, and three deadlines hit this week.

GPT-5.6 is callable today, and Luna's price depends on which OpenAI page you read

As covered yesterday, OpenAI confirmed Tuesday that GPT-5.6 Sol, Terra, and Luna would go public July 9. They did: OpenAI posted on X on July 8 that the three models launch publicly Thursday, and the rollout is now expanding across the API, Codex, and ChatGPT. Expect staged access by product and account tier rather than everything at once.

The pricing story has a wrinkle nobody flagged on the way in. OpenAI's launch blog post and Help Center article both list Luna at $1 input / $6 output per million tokens. OpenAI's developer docs models page, the page you actually build against, lists the third tier at $0.75 input / $4.50 output, with a 400K-token context window (versus 1M for Sol and Terra) and a "Faster" latency tag. That is 25% cheaper on both input and output than the number in the launch materials, and it is the only place a context window for Luna appears at all.

Sol and Terra are consistent across both surfaces: Sol at $5 / $30 with a 1M context and a Dec 1, 2025 knowledge cutoff; Terra at $2.50 / $15 with a 1M context and an Aug 31, 2025 cutoff. Terra is the one OpenAI is selling hardest, pitched as GPT-5.5-competitive performance at half the cost. Sol Fast mode runs $12.50 / $75, 2.5x the standard rate, at up to 750 tokens per second.

Until OpenAI reconciles the two pages, treat the developer docs as the billing source of truth and verify Luna's rate in your own dashboard before locking in cost projections. The cache fine print, which carried over from preview, is unchanged: explicit cache breakpoints, a 30-minute minimum cache life, cache writes at 1.25x the uncached input rate, and cache reads at a 90% discount. For any app sending repeated system prompts or long shared context, the 90% read discount still makes caching heavily net positive; you just instrument it explicitly now.

GPT-Live voice models shipped the day before GPT-5.6

OpenAI also released GPT-Live on July 8, a full-duplex voice model family that replaces Advanced Voice Mode in ChatGPT. Two versions ship first: GPT-Live-1, the default for Go, Plus, and Pro subscribers, and GPT-Live-1 mini, the default for free users. Full-duplex means the model listens and speaks at the same time, so you can interrupt mid-sentence, and it powers live translation and a "Hey Chat" wake word, per TechCrunch.

For the pricing beat the catch is access. GPT-Live is a ChatGPT consumer feature for now, with an API described as coming soon and no published per-minute or per-token rate yet. OpenAI's separate realtime API models, gpt-realtime-2.1 and gpt-realtime-2.1-mini, remain the billed voice endpoints if you are building. One detail worth knowing: GPT-Live delegates deeper reasoning to GPT-5.5 in the background, not GPT-5.6, so even on launch day your voice surface and your newest text model are not the same thing.

Current prices, July 9

Every line links to the official pricing or docs page so you can verify at the source. Prices are per 1M input / 1M output tokens unless noted.

• GPT-5.6 Sol $5 / $30, 1M context. Terra $2.50 / $15, 1M context. Luna $0.75 / $4.50 with a 400K context per the developer docs; $1 / $6 per the launch post. GPT-5.5 $5 / $30, 272K context. Full card at OpenAI pricing.
• Claude Fable 5 $10 / $50. Opus 4.8 $5 / $25. Sonnet 5 $2 / $10 intro through Aug 31, then $3 / $15 on Sep 1. Haiku 4.5 $1 / $5. Fable 5, Mythos 5, Opus 4.7 and later, and Sonnet 5 use a tokenizer that produces roughly 30% more tokens for the same text, so headline per-token rates understate real cost. Anthropic pricing.
• Gemini 3.5 Flash $1.50 / $9 (standard), with Batch and Flex tiers at $0.75 / $4.50. 3.1 Flash-Lite $0.25 / $1.50. Gemini 3.5 Pro is not yet generally available. Google pricing.
• Grok 4.3 $1.25 / $2.50, 1M context. Grok 4.5 $2 / $6, 500K context. Batch API discounts vary by model. xAI pricing.
• Mistral Large 3 $0.50 / $1.50, open-weight under Apache 2.0. Mistral Medium 3.5 $1.50 / $7.50. Magistral Medium $2 / $5. Mistral pricing.
• DeepSeek V4-Pro $0.435 / $0.87 (cache-miss input / output), cache-hit input $0.003625, 1M context. V4-Flash is cheaper. A mid-July launch brings 2x peak-hour pricing during 9am to 12pm and 2pm to 6pm Beijing time. DeepSeek pricing.

Deadlines and tracking

• DeepSeek legacy names retire July 24, 15 days out. deepseek-chat and deepseek-reasoner go dark at 15:59 UTC; route to v4-flash non-thinking and thinking modes until then. V4's official launch is still "mid-July" with no exact date confirmed. Deprecation notice.
• Claude Fable 5 subscription access ends July 12, 3 days out. After 11:59pm PT it leaves Pro, Max, Team, and Enterprise and bills at $10 / $50 usage credits. Extension report.
• Sonnet 5's intro price ends August 31. Sep 1 jumps to $3 / $15, and the roughly 30% tokenizer inflation makes the effective increase larger than the headline suggests. Anthropic pricing.
• Gemini 3.5 Pro targets July 17, still unconfirmed by Google and now attributed to an architectural rebuild. No published benchmarks or finalized pricing; a standard tier around $1.25 / $10 has been reported but not official. Reporting.
• Mistral's next frontier open-weight model opened early access this month, per CEO Arthur Mensch, with general availability later this summer. No name, specs, or price yet; Apache 2.0 is expected. Early access.
• Meta Muse Image shipped July 7 across Meta AI, Instagram, and WhatsApp, with Muse Video marked coming soon. Consumer surface only, no API pricing disclosed. Meta launch.

Want the next one?

Every new AI Releases & Pricing issue by email. One tap to unsubscribe.