Skip to content
NewsModel release

Meta Muse Spark 1.3 Launches at $1.25 Per Million Input Tokens with 1M Context Window and Reasoning

· by Pondero Newsdesk

The short version

Meta released Muse Spark 1.3 on September 2, 2026, pricing the multimodal reasoning model at $1.25 per million input tokens and $4.25 per million output tokens. The model places 13th on the Artificial Analysis Intelligence Index with a score of 48, well above the median of 24 for comparable models.

Meta Muse Spark 1.3 Launches at $1.25 Per Million Input Tokens with 1M Context Window and Reasoning

At $1.25 per million input tokens, Muse Spark 1.3 arrives at roughly half the input price of several competing frontier reasoning models, while carrying a 1-million-token context window and multimodal inputs out of the gate. Meta made the model available on September 2, 2026 through its Muse Code coding tool and the Meta Model API.

What happened

Meta released Muse Spark 1.3 on September 2, 2026, per the Meta AI Research blog. The model is available in Muse Code and through the Meta Model API at dev.meta.ai. Pricing, per Artificial Analysis, is $1.25 per million input tokens, $4.25 per million output tokens, and an 88% cache discount on prompt cache hits.

The model handles text, image, and video inputs and produces text output. Its context window is 1 million tokens. Muse Spark 1.3 is a reasoning model, meaning it generates extended chain-of-thought before returning a final answer.

On the Artificial Analysis Intelligence Index, Muse Spark 1.3 scores 48, placing it 13th among 202 evaluated models, per Artificial Analysis. The median score for comparable models in its class is 24, putting it well above the midpoint of the field. Output speed reached 232 tokens per second in Artificial Analysis testing, ranking it 7th for speed across the same cohort.

Meta benchmarked the model against its predecessor Muse Spark 1.2, GPT 5.6 Sol, and Opus 5 across agentic, coding, instruction-following, and long-context tasks, per the Meta AI Research blog. In internal testing by Meta engineers, Muse Spark 1.3 required roughly 20% fewer tool calls and roughly 25% fewer tokens than Muse Spark 1.2 to complete the same coding workflows.

Meta also described improved behavior for long-horizon agentic tasks: the model asks clarifying questions when a prompt is ambiguous, flags gaps in its own plan rather than proceeding silently, and confirms before taking actions it classifies as consequential. On safety, Meta said the model shows stronger resistance to adversarial inputs and prompt injection attempts than its predecessor.

A planned open weights release for Muse Spark appears in the "Looking Forward" section of the announcement. No timeline was given.

Why it matters

The $1.25 input price is the headline number for API teams running cost-sensitive workloads. At batch API rates, input costs could fall further. The 88% prompt cache discount is among the largest published cache discounts currently available at the frontier level; any workload with substantial repeated context will see effective input costs drop well below the listed rate on cache hits.

The 1-million-token context window opens Muse Spark 1.3 to document-analysis, retrieval, and long-session use cases without chunking. Video input support broadens the model's scope beyond text and image agents.

For teams already using Muse Code as their primary coding environment, the upgrade to Muse Spark 1.3 arrives automatically. API users who built on earlier Muse Spark endpoints should check their model ID references at dev.meta.ai to confirm they are routing to 1.3.

What to watch next

The planned open weights release would let teams run Muse Spark locally or on their own infrastructure, removing the per-token pricing exposure entirely. Watch Meta AI Research for a timing announcement.

Artificial Analysis updates its Intelligence Index scores as it adds new evaluations; Muse Spark 1.3's current 13th-place ranking will shift as the field and the benchmark suite both grow.

Sources