AI news July 11 2026: GPT-5.6 Sol lands at half Fable's price, Meta enters coding, $412B in VC

OpenAI launches GPT-5.6 Sol at $5 per million input tokens, half the price of Anthropic's Fable 5. The flagship model of the new GPT-5.6 family tops the Coding Agent Index at 80 (2.8 points above Fable 5), and Sam Altman tells CNBC that Sol is 54% more token-efficient on agentic coding tasks.

SpaceXAI (xAI's new name after Monday's rebrand) ships Grok 4.5, which Musk calls an "Opus-class model." Meta jumps into AI coding with Muse Spark 1.1 at $1.25 per million tokens. Anthropic extends free Fable 5 access through July 12. And PitchBook reveals that US venture capital hit $412.7 billion in 6 months, with 86% going to AI.

The rebrand was predictable. SpaceX acquired xAI in February 2026, putting space, AI, and social media under one roof. After SpaceX's blockbuster IPO in June, Musk decided xAI no longer needed to exist as an independent entity (Business Insider). Gizmodo titled it "xAI Is Dead. Long Live SpaceXAI" (Gizmodo).

Grok 4.5 targets agentic and coding workflows. SpaceXAI internally compares it to Claude Opus 4.8 and GPT-5.5 (Gizmodo). Cursor's founder was among the early testers. Forbes reports the model is competitive on financial tasks and autonomous workflows, and that Musk calls it the company's "strongest model ever."

The rebrand is Musk at his most predictable: absorb everything under a single brand. Grok 4.5 at $2 per million tokens is competitive on paper. But Grok has never stood out for the reliability of its code responses. A self-proclaimed "Opus-class" model on X, without solid independent benchmarks, is marketing first and foremost.

But while Musk consolidates brands, Meta shows up out of nowhere on a battlefield nobody expected them on.

Muse Spark 1.1 was built by Meta's Superintelligence Labs. Unlike Llama, which remains an open-weight generalist model, Muse Spark is a closed API product, built for agentic use (the ability for a model to orchestrate complex tasks autonomously, like detecting bugs, planning code migrations, or chaining actions without human intervention). The Verge notes the model is designed to "work well with all of the most popular harnesses that developers use today." CNET confirms it supports advanced bug detection, large-scale code migrations, and multimodal reasoning.

The $1.25 per million input tokens rate is simply the lowest of any coding model in this category. Fable 5 costs $10, Sol $5, Grok 4.5 sits at $2, and Muse Spark undercuts everything at $1.25. Meta calls its pricing "aggressive and attractive," which is an understatement. TechCrunch notes that the popularity of OpenClaw (an open-source tool for managing AI agents) in H1 2026 has accelerated the coding agent hype, and Meta wants its share.

The real signal here is not the model's performance (which cannot be seriously compared to competitors yet). It is the fact that Meta, a social media giant, is entering the AI coding race with pricing that puts pressure on the entire market. At $1.25 per million tokens, cost will never be a differentiator again. Code quality will be what separates the contenders. For Waku and my projects, I will not switch to Muse Spark tomorrow. But the signal is clear: coding model costs will keep falling, and that is good news for every indie dev.

Meanwhile, Anthropic is playing the clock on Fable 5.

The extension follows the lifting of export controls by the US Department of Commerce. Restrictions on Fable 5 and Mythos 5 were removed after Anthropic worked with the government (AOL). I covered Fable 5's rocky return last week. Result: Anthropic restores Fable 5 across all platforms, including Claude.ai, Claude Code, and Claude Cowork.

In parallel, Anthropic announces that Claude Cowork is available on mobile and web (MacRumors). You can now start a work session on desktop and pick it up on your phone. Cowork usage limits are doubled through early August.

Five extra days of free Fable 5 access is a commercial gesture, not a strategy. The real question is what happens after July 12. At $10 per million input tokens (vs. $5 for Sol and $1.25 for Muse Spark), Fable 5 is the most expensive model on the market. And while Anthropic stretches the free window, OpenAI brings out the big guns. I use it daily on Waku with Claude Code, and the credit cost question will quickly become central in choosing a default model.

All that was the appetizer. The main course arrived Thursday.

This is OpenAI's most anticipated release since GPT-5. The GPT-5.6 family was unveiled in late June, but the Trump administration requested limited access for a government review period. On July 8, 9to5Mac announced the green light, and Sam Altman confirmed on X that the public rollout would happen Thursday July 9 (CNET).

Sol is the flagship, built for complex reasoning, end-to-end technical execution, and design. Terra is the everyday workhorse, balanced between power and cost. Luna is optimized for speed. Axios notes that an "ultra" mode within Sol lets the system "work even harder on tasks" and delegate work to other models in the family. Forbes sums it up: Sol is the brain, Terra the right hand, Luna the sprinter.

OpenAI claims Sol is its "best coding model ever" and directly compares it to Anthropic's Fable 5. The key figures:

Coding Agent Index (Artificial Analysis): Sol 80, Fable 5 77.2. Terminal-Bench 2.1: Sol 91.9%. Token efficiency: 54% more efficient on agentic coding tasks, per Altman's statement to CNBC. AI Intelligence Index: Sol 59 points, Fable 5 60 points (evaluated with fallback), per TrendingTopics.

The picture is not black and white. Axios reports that investor Matt Shumer wrote on X: "It's an amazing model, but for almost every task I tested, Fable was quite a bit better." On the other hand, Theo Browne, CEO of T3 Chat, states: "gpt-5.6-sol is world-leading in computer use. It made me use it 100x more" (Axios). And Pietro Schirano, CEO of MagicPath AI: "I've been testing it for months and, without exaggeration, it's the best model I've ever used" (Axios).

On Google Android Bench, Fable 5 still leads with 84.5 points vs. 80.2 for GPT-5.5 (Sol has not been evaluated on this benchmark yet). The emerging consensus: Sol dominates coding and value for money, Fable 5 retains the edge in pure reasoning and general intelligence.

The same day, OpenAI launches ChatGPT Work, an agent that combines ChatGPT's conversational abilities with Codex's coding capabilities. The Verge describes it as the realization of OpenAI's "super app" vision. The tool integrates with Slack, Teams, Google Drive, SharePoint, Gmail, calendars, and CRMs through a unified plugin directory (Engadget). It can produce documents, spreadsheets, presentations, reports, and even mini web apps (Reuters). Ars Technica notes that unlike last year's Agent mode, ChatGPT Work can maintain long sessions without stopping after a few minutes.

Sol at $5 per million input tokens is the best quality-to-price ratio on the market right now. The Coding Agent Index score of 80 is a real signal. But I am not ready to ditch Claude Code. The thing is, benchmarks measure isolated tasks. In real conditions, on an existing repo with context and dependencies, the difference between Sol and Fable 5 comes down to details benchmarks do not capture: project context understanding, the ability to not break what already exists, refactoring suggestion quality. On Waku, I work with Claude Code and Anthropic's models daily (I wrote about it here). Sol at half the price is tempting, but cost per token means nothing if the model needs twice as many attempts to get the right result. The real metric is cost per completed task.

And then there is ChatGPT Work. The idea of combining chat, code, and office tools in a single interface is exactly where every lab is converging. Claude Cowork does the same thing at Anthropic. The question is no longer "who has the best model" but "who builds the best product around the model." That is a paradigm shift I had already identified in the article on agent architecture.

The number that says it all: Anthropic raised $65 billion in a single round, bringing its post-money valuation to $965 billion. Three mega-funds (Andreessen Horowitz, Thrive Capital, and Founders Fund) raised $34.8 billion on their own, nearly half of all new VC capital for the period. TechCrunch counts roughly 90 new unicorns (startups valued at over $1 billion) created in H1, including Avoca (AI agents for customer support), Core Automation, and Omni Analytics.

On the spending side, Business Insider reports that Amazon, Alphabet, Meta, and Microsoft are on track for $725 billion in AI capex in 2026. By the end of the decade, the combined AI spending of these four giants is expected to exceed the GDP of countries like Brazil.

These numbers explain everything happening this week. When 86% of VC capital goes to AI, every lab has to ship a new model every two weeks to justify its valuation. OpenAI ships GPT-5.6, Meta launches Muse Spark, SpaceXAI drops Grok 4.5, Anthropic extends Fable 5. It is an arms race funded by amounts with no historical precedent. $412 billion in 6 months is massive, and the question is not "will the bubble burst" but "how long can this pace hold."

GPT-Live: OpenAI's full-duplex voice. OpenAI launches GPT-Live-1 and GPT-Live-1 mini, voice models that can listen and speak simultaneously (full-duplex architecture, meaning the model no longer needs to wait for the user to finish talking). Available on iOS, Android, and web. VentureBeat notes it is the third generation of ChatGPT's voice tech in two years.

Character.AI enters the microdrama business. The AI chatbot platform launches interactive micro-series where viewers can converse with AI characters before, during, and after episodes (Lapaas Voice). A crossover between streaming and conversational AI.

Google updates Android Bench. Fable 5 leads with 84.5 points, 4 points ahead of GPT-5.5 at 80.2. 9to5Google notes Google is changing its testing methodology to integrate more realistic Android code evaluations.

AI Governance Week in Geneva. Early July, Geneva hosts the largest concentration of international AI governance events ever organized. UN Secretary-General Antonio Guterres calls for a global regulatory framework (IPU).

This week alone packs more model releases than some entire quarters of 2025. OpenAI, Meta, SpaceXAI, and Anthropic all played their hand within 72 hours. Business Insider called these 72 hours a "whirlwind" of rival announcements. The pattern is clear: every lab watches the others' announcements and synchronizes its own releases to avoid being eclipsed.

The real winner this week is the developer. GPT-5.6 Sol at $5 per million tokens, Muse Spark at $1.25, Grok 4.5 at $2. Six months ago, these prices would have been unthinkable for models of this caliber. Competition drives prices down and quality up, which is exactly what indie developers need. The cost per completed task keeps falling.

Have you tested Sol yet? Staying on Fable 5 or thinking about switching? Hit me up on Twitter/X or drop a comment.

Alex