Meta has entered the paid AI market with aggressive pricing designed to challenge Anthropic's Claude and OpenAI's ChatGPT. On July 9, the company unveiled Muse Spark 1.1, its first proprietary, closed-source AI model available through a paid API. The pricing strategy is striking: Meta charges $1.25 per million input tokens and $4.25 per million output tokens, undercutting every major competitor on both metrics.

This launch represents a dramatic strategic reversal for Meta, which built its AI reputation on freely distributed open-source Llama models. The shift reflects new Chief AI Officer Alexandr Wang's mandate to transform Meta's AI research into a revenue-generating business. Wang, a 28-year-old former co-founder and CEO of Scale AI, was recruited in June 2025 when Meta acquired a 49% stake in Scale for $14.3 billion.

How Does Meta's Pricing Compare to Competitors?

The pricing gap between Meta and its rivals is substantial, particularly for developers running high-volume workloads. Here's what the numbers reveal:

• Input Token Cost: Meta's $1.25 rate is 37% cheaper than Anthropic's Claude Sonnet 5 introductory pricing of $2, and 75% cheaper than both Anthropic's Opus 4.8 and OpenAI's GPT-5.5, which charge $5 per million input tokens.
• Output Token Cost: Meta's $4.25 output rate is 58% below Sonnet 5's introductory $10, 83% below Opus 4.8's $25, and 86% below GPT-5.5's $30 per million output tokens.
• Real-World Impact: A developer running the same task across all four models would pay Meta's rate for less than one-third of what GPT-5.5 charges on output alone, according to pricing analysis.

The gap narrows somewhat after August 31, when Sonnet 5's introductory pricing expires and shifts to a standard $3 per million input tokens and $15 per million output tokens. However, Meta still undercuts both Claude and ChatGPT on both metrics even at standard rates.

What Can Muse Spark 1.1 Actually Do?

Muse Spark 1.1 is purpose-built for coding and agentic AI tasks, meaning it's designed to handle complex workflows where AI systems coordinate multiple actions autonomously. The model supports a one-million-token context window, which translates to roughly 750,000 words of text. This capacity allows developers to process entire codebases in a single session without breaking up their work.

According to Meta, the model handles bug diagnosis, feature implementation in enterprise systems, large-scale code migrations, tool calling, function calling, and computer-use workflows. It can also coordinate parallel subagent systems, functioning as both a primary agent and a subagent for complex tasks.

Meta claims Muse Spark 1.1 rivals GPT-5.5 and Opus 4.8 on agentic evaluation benchmarks, achieving top positions on specialized tests like MedScribe, TaxEval, and Harvey's Legal Agent Bench. The company also states the model is "10x cheaper and twice as fast" than competing solutions. However, Meta has not published independent benchmark scores against either rival, so these claims remain unverified by third parties.

The original Muse Spark 1.0, released in April, had notable coding weaknesses. It scored 59.0 on coding benchmarks compared to 80.8 for Claude and 75.1 for OpenAI's GPT-5.4, according to the Artificial Analysis Intelligence Index. The 1.1 upgrade is specifically targeted at closing that performance gap.

Who Has Access, and What Are the Limitations?

Access to Muse Spark 1.1 remains limited in its early stages. The public preview is available only to U.S.-based developers and requires joining a waitlist. New accounts receive $20 in free credits before billing begins. The model is not yet available on third-party marketplaces like OpenRouter, which limits distribution compared to Claude and ChatGPT.

Early API partners include Replit, Cline, and Box. Replit CEO Amjad Masad highlighted the million-token context window and OpenAI-compatible API format as key advantages. Cline CEO Saoud Rizwan pointed to the model's tool usage capabilities and pricing advantages for scaled coding workloads.

The OpenAI-compatible API format is significant because it lowers the barrier for developers to switch from competing platforms. Developers already using OpenAI's API can integrate Muse Spark 1.1 with minimal code changes.

How to Evaluate Muse Spark 1.1 for Your Development Needs

• Cost Analysis: Calculate your expected token usage across input and output to determine actual savings compared to Claude and ChatGPT, since pricing advantage matters most for high-volume workloads.
• Benchmark Review: Compare Muse Spark 1.1's performance on your specific coding tasks against Claude and GPT-5.5, since Meta's claims remain unverified by independent third parties.
• Integration Testing: Test the OpenAI-compatible API format with your existing development stack to ensure minimal migration effort if you decide to switch from a competing platform.
• Waitlist Registration: Join the public preview waitlist if you're a U.S.-based developer interested in accessing the $20 in free credits before standard billing begins.

Why Is Meta Making This Strategic Shift?

Meta's move from open-source to proprietary, paid models marks a philosophical reversal driven by competitive necessity. The company spent years evangelizing open AI development and distributing Llama models freely, arguing that open-source AI benefited the broader ecosystem and countered the dominance of closed-model competitors.

However, Llama 4 underperformed in the market, and Claude became the default model for AI-assisted coding tools in 2025 and 2026. Meta's new strategy under Wang aims to establish the company as a credible alternative to Anthropic and OpenAI in the enterprise AI market. The rapid release cadence supports this ambition. Beyond Muse Spark 1.1, Meta released Muse Image, its first image-generation model, earlier in the week.

"Very aggressive and attractive," said Alexandr Wang, describing the pricing strategy to CNBC.

Alexandr Wang, Chief AI Officer at Meta

The AI coding tools market has become one of the highest-growth segments in enterprise software. By entering with a paid API and aggressive pricing, Meta is attempting to capture market share from Anthropic and OpenAI while leveraging its distribution advantage across billions of users on Facebook, Instagram, WhatsApp, and Ray-Ban smart glasses.

Muse Spark 1.1 will replace existing Llama models powering chatbots across Meta's consumer apps, giving the new model immediate exposure to a massive user base. Whether this distribution advantage translates to developer adoption remains to be seen, as real-world coding performance and reliability matter more than pricing alone.