AI Releases & Pricing

July 10, 2026

Muse Spark 1.1 Is Meta's First Paid Model, Priced Below Grok 4.5 at $1.25/$4.25

▶ Listen 🎧 Podcast app

Meta opened its first paid API at Haiku-tier pricing with a million-token context. Grok 4.5 shipped the same week at $2/$6 with half its predecessor's context. OpenAI reconciled the Luna price gap, and the Fable 5 subscription cliff is 48 hours out.

Meta's first paid model: Muse Spark 1.1, $1.25/$4.25, 1M context

Meta has never sold a model before. That changed July 9, when Meta Superintelligence Labs shipped Muse Spark 1.1 and opened the Meta Model API in public preview. The pricing is aggressive: $1.25 per million input tokens, $0.15 per million cached input, and $4.25 per million output. New accounts get $20 in free credits.

The context window is 1 million tokens, actively managed. The model compacts earlier steps, retrieves from deep in a session, and keeps intermediate state for later work. It takes text, image, video, and PDF input, supports MCP servers and custom tools zero-shot, and ships with an OpenAI-compatible and Anthropic-compatible API surface. Point an existing client at api.meta.ai/v1, set the model to muse-spark-1.1, and keep your code.

The fine print matters here. Muse Spark 1.1 is a reasoning model, and its internal thinking tokens are billed at the output rate, so a heavy reasoning workload costs more than the headline per-token price suggests. Web search grounding is $2.50 per 1,000 queries. The preview is US-only, new users join a waitlist, and Meta is deliberately keeping it off third-party platforms like OpenRouter for now. This is also Meta's first closed-source, paid model, a sharp break from its open-weights Llama tradition.

At $1.25/$4.25, Muse Spark 1.1 sits in the same tier as Claude Haiku 4.5 at $1/$5 rather than competing directly with Opus or Sol. The 1M context at that price is the differentiator. For sustained agentic coding sessions that fit in a single context instead of chunking and re-prompting, the per-task cost can beat models that look cheaper on a per-token basis.

Grok 4.5: $2/$6, 500K context, free for now

The same week, SpaceXAI launched Grok 4.5 at $2 per million input tokens and $6 per million output tokens, per the official pricing page. It is available now in Grok Build, in Cursor on all plans, and from the SpaceXAI console. Usage is free for a limited time in Grok Build and Cursor.

The context window is 500K tokens. That is half the 1M context of grok-4.3, which is still listed at $1.25/$2.50 on the same pricing page. The newer model is pricier and has less context than its predecessor. SpaceXAI is betting that token efficiency closes the gap: the company claims roughly 2x the token efficiency of comparable leading models, served at 80 tokens per second. EU availability is expected in mid-July, and Grok 4.5 is not yet available in the EU in any SpaceXAI product or the API console.

One pricing detail worth checking before committing volume: the Batch API gives 20% off standard rates for grok-4.3 and the grok-4.20 variants, but grok-4.5 is not on the batch-discount list. Priority Processing is available at a 2x premium over standard rates for all token types.

OpenAI reconciled the Luna price gap

Yesterday this feed flagged a discrepancy on OpenAI's own pages: the developer docs models page listed GPT-5.6 Luna at $0.75/$4.50 with a 400K context window, while the launch post and Help Center said $1/$6 with no context figure. That gap is gone. The live models page now lists Luna at $1 input and $6 output per million tokens, matching the launch materials, and gives Luna a 1.05M context window, the same as Sol and Terra. All three tiers now show 1.05M context and 128K max output.

The cache terms are consistent across all three: cache writes billed at 1.25x the uncached input rate, cache reads at a 90% discount, and a 30-minute minimum cache life. Sol is $5/$30, Terra is $2.50/$15, Luna is $1/$6. If you built against the earlier dev-docs numbers, re-pull the page, because both the price and the context window moved.

Tracking

Current prices (verified July 10, 2026)

Prices per 1M input/output tokens, linked to each provider's official pricing page.

Want the next one?

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