Meta's Muse Spark 1.2: A Genuinely Impressive Mid-Tier Model at a Bargain Price

Meta just dropped another update to its Muse family, and this one is worth paying attention to. Muse Spark 1.2 launched on August 5, 2026, and it lands an overall score of 9 out of 10 in our evaluation, backed by strong benchmark numbers and pricing that undercuts most of its peers.

## What Is Muse Spark 1.2 Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1, with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows, built by significantly scaling up training compute on coding tasks while expanding training environment diversity. The model also maintains its strength in other key areas like general agents.

The release didn't come alone. Meta also released Muse Code in beta alongside it, a terminal coding agent that takes on complete software engineering tasks across large repos, planning changes, writing code, and validating the results, powered by Muse Spark 1.2.

One of the more interesting technical details from Meta's own writeup is how the model trained itself. Muse Spark 1.1 was used to generate challenging coding environments and instruction-following templates, then graded candidate solutions on how well they satisfied those requirements, producing a scalable training dataset for Muse Spark 1.2, a self-improvement loop that helped the new model follow complex instructions more precisely than its predecessor.

Meta also pushed the model on extreme long-horizon tasks. The team tested the model's ability to iteratively optimize GPU kernels over 1,000+ tool calls, running up to 24 hours, using Muse Code's agentic coding environment to write, compile, profile, and progressively improve kernel performance relative to a provided baseline, benchmarked on KDA and MLA kernels for NVIDIA Hopper GPUs.

## The Numbers That Matter Here is how Muse Spark 1.2 stacks up on the core specs: | Metric | Muse Spark 1.2 | |---|---| | Overall Score | 9 out of 10 | | Input Price | $1.25 per 1M tokens | | Output Price | $4.25 per 1M tokens | | Artificial Analysis Intelligence Index | 54 | | Context Window | 1M tokens | | Release Date | August 5, 2026 | | Modalities | Text, video, image, audio |

That Intelligence Index score of 54 is a meaningful jump for the Muse Spark line. For context, the original Muse Spark scored 43 on the index at launch in April, and Muse Spark 1.1 followed three months later at 51, tied at the top of the mid-tier pack with GLM-5.2 (max) and GPT-5.6 Luna (max), also at 51, and sitting three points behind Grok 4.5 (high) at 54. With 1.2 now scoring 54, Meta has essentially closed that gap and pulled level with Grok 4.5, a notable achievement given how competitive the mid-tier has become.

## Why the Pricing Stands Out Meta has kept its API pricing steady across the Muse Spark updates, and that consistency is part of what makes this model so compelling. The $1.25 input and $4.25 output pricing was already competitive when 1.1 launched. Reuters reported that the company charges $1.25 per million input tokens and $4.25 per million output tokens, putting it in line with, albeit slightly above, Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna.

That combination of low cost and rising intelligence score is exactly why this model earns a high overall score. When Muse Spark 1.1 was evaluated at the same price point, analysts noted that the cost worked out to roughly $0.26 per Intelligence Index task on Meta's public pricing, undercutting GLM-5.2 and running well below several of the other models it was tied with on raw intelligence score, a combination Meta hadn't managed before. With Muse Spark 1.2 pushing the intelligence score even higher at the same price, that value proposition only gets stronger.

## Not Quite the Absolute Frontier, and That's Fine To be clear, Muse Spark 1.2 is not trying to dethrone the very top of the leaderboard. The leading edge of the field is still occupied by models like Claude Fable 5, GPT-5.6 Sol, and Claude Opus 4.8, all of which score well above 54 on the Intelligence Index. But Muse Spark 1.2 was never positioned as an attempt to out-muscle those flagship, high-cost models. Instead, it competes in the crowded medium tier where price and practical performance matter more than chasing the last few points on a benchmark leaderboard.

## Coding Is the Real Story Where Muse Spark 1.2 seems to differentiate itself most is coding. Meta paired the model release with Muse Code specifically because the earlier Muse Spark releases had gaps in this area. Muse Spark competed with products from Google and OpenAI, though there were gaps in areas like coding workflows and complex multistep agentic tasks compared with models from outside labs, per the company. The 1.2 update looks like Meta's direct answer to that criticism, doubling down on training compute specifically for coding and shipping a dedicated coding agent to go with it.

## The Bigger Picture Muse Spark 1.2 is the latest step in a rapid cadence of releases from Meta Superintelligence Labs. The original Muse Spark arrived in April 2026 as Meta's re-entry into the frontier race after roughly a year of relative quiet, following the company's reorganization of its AI efforts under Meta Superintelligence Labs. Three months later, Muse Spark 1.1 brought a meaningful jump in intelligence score alongside Meta's first paid API access. Now, roughly a month after that, Muse Spark 1.2 arrives with a sharper focus on coding and another bump in overall capability.

For developers and businesses looking for a model that delivers strong reasoning, multimodal understanding across text, video, image, and audio, a full 1M token context window, and genuinely competitive pricing, Muse Spark 1.2 is hard to ignore. It may not be sitting at the very top of the leaderboard, but a 9 out of 10 overall score paired with pricing this aggressive makes it one of the smartest choices in its class right now.