AI Releases & Pricing

July 8, 2026

GPT-5.6 Sol, Terra, and Luna Launch July 9. Terra Is Half the Price of GPT-5.5.

▶ Listen 🎧 Podcast app

Commerce cleared the last regulatory hurdle, preview access opened globally, and Anthropic extended Fable 5's free window to July 12 in response. Plus: Sonnet 5 has a September price jump hiding in the fine print, and the token market is splitting in two.

GPT-5.6 goes public tomorrow, with full pricing confirmed

OpenAI confirmed today that GPT-5.6 Sol, Terra, and Luna will be publicly available starting Thursday, July 9. Sam Altman posted on X: "GPT-5.6 sol launches thursday! happy building." The U.S. Department of Commerce approved a broad launch after additional testing and meetings with government agencies, Axios reported on Tuesday. Preview access is now global, no longer limited to U.S.-only vetted partners.

This is the resolution of the story covered here yesterday, when the Polymarket-fueled July 7 GA rumor did not land and GPT-5.6 remained government-gated. One day later, the date is real.

The three-model lineup and pricing, per OpenAI's announcement:

One fine-print detail: OpenAI introduced more predictable prompt caching for the 5.6 series, including support for explicit cache breakpoints and a 30-minute minimum cache life. For agent loops that resend the same context every turn, that minimum cache life matters more than the headline rate. OpenAI also teased an "Ultra" mode where multiple AI agents collaborate on complex requests, but specific pricing for Ultra was not announced.

Anthropic extends Fable 5 to July 12, five days past the original cutoff

Hours before Fable 5 was supposed to leave subscription plans on July 7, Anthropic announced on X: "We're extending access to Claude Fable 5 on all paid plans through July 12." The extension came after backlash online over the early cutoff.

Neowin reported that the timing was not a coincidence: "Anticipating the GPT-5.6 series launch, in order to maintain mindshare with developers, Anthropic today announced that it is expanding promotional access to Claude Fable 5." Yesterday's issue reported the cutoff as final. It is not.

Nothing else changed. Pro, Max, Team, and premium Enterprise subscribers can still use Fable 5 for up to 50% of their weekly usage limit at no extra cost. After July 12 at 11:59 PM PT, the model shifts to usage credits at $10 per million input / $50 per million output, the highest published rate for any generally available Anthropic model. Anthropic has said the shift is a temporary capacity measure and plans to restore Fable 5 to standard subscription plans when compute allows, with no timeline given.

Sonnet 5's intro pricing ends August 31, and the tokenizer makes the increase larger than it looks

The Anthropic pricing page contains a time bomb. Claude Sonnet 5's introductory rate of $2 / $10 per million tokens runs only through August 31, 2026. Starting September 1, the standard rate of $3 / $15 takes effect, a 50% increase on both input and output.

The tokenizer makes the real increase larger. Anthropic's own docs confirm that Sonnet 5 uses a newer tokenizer that produces approximately 30% more tokens for the same text. So the effective cost today is about $2.60 / $13 per million tokens of equivalent text, and from September 1 it jumps to about $3.90 / $19.50. Sonnet 4.6, with the old tokenizer, costs $3 / $15. Once the intro window closes, Sonnet 5 will cost more than the model it replaced, not less.

Kimi K2.7 reaches Copilot Business and Enterprise

GitHub announced on July 7 that Kimi K2.7 Code, the first open-weight model in the Copilot model picker, is now available for Copilot Business and Copilot Enterprise plans. It was already available to Pro, Pro+, and Max subscribers since July 1.

The model is off by default for Business and Enterprise. Plan administrators must enable the Kimi K2.7 Code policy in Copilot settings before anyone in the organization can select it. It is billed at provider list pricing under usage-based billing, at approximately $0.95 / $4.00 per million tokens, roughly 5x cheaper than Claude Sonnet 5 or GPT-5.5 for coding workflows. The weights are released under a Modified MIT license through Hugging Face and Ollama, meaning you can pull them, run them on your own hardware, and audit them.

The token market is splitting in two

The Register reported today on a structural shift in AI pricing. GPT-4-class model output cost about $20 per million tokens in late 2022. Today, equivalent capability costs about $0.40, a 55x decline in less than four years. But frontier model prices are moving the opposite direction. OpenAI doubled GPT-5.5 to $5 / $30. Google's Gemini Flash 3.5 arrived three to six times more expensive than the model it replaced.

Open-weight models are closing the quality gap. Larridin's CTO Ameya Kanitkar told The Register that Kimi 2.6/2.7 and GLM 5.2 "are almost at parity with Opus 4.7 or 4.8. And they are 10x cheaper in theory, or about 5x cheaper in practice." They consume more tokens and run slower, but the per-token cost is low enough to absorb that.

The spending pattern is shifting too. Kanitkar said companies are now spending between 10 and 20% of their labor cost on tokens. Larridin found that 15 to 30% of AI users account for more than 50% of total spend, and there is an inflection point at about 35 to 40% of client spending where burning more tokens stops boosting productivity. Using that point as a cap can cut costs 40% without changing anything else. Even so, enterprises still direct almost half of their AI spending toward Anthropic's Opus because it handles complex engineering and reasoning tasks well.

Tracking

Current prices: what every comparable model costs right now

All prices are per 1 million tokens, input / output, at standard (non-batch) rates. Linked to the official pricing page for each. Verified July 8, 2026.

OpenAI (pricing page):

Anthropic (pricing page):

Google (pricing page):

xAI (pricing page):

DeepSeek (pricing page):

Open-weight:

Want the next one?

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