Meta Platforms, Inc. (NASDAQ: META) launched a new AI coding agent called Muse Code on August 5, pricing it well below rival products from Anthropic and OpenAI in a direct push to win over cost-conscious developers.
Muse Code is structured around two pricing tiers, with the most aggressive option set at just 20 cents per million output tokens for users willing to share usage feedback with Meta.
That discounted tier aligns with pricing from China’s DeepSeek and cuts beneath even OpenAI’s older, reduced-cost model offerings, positioning Meta as the budget-friendly option in a crowded market.
Meta AI chief Alexandr Wang described the value proposition plainly, saying the pricing “can be an incredibly good option for a lot of workflows, especially from a cost perspective.”
The launch carries significant strategic weight for Meta, whose shares fell 10% in the week before Muse Code’s release after CEO Mark Zuckerberg offered investors little new detail about the company’s cloud-computing plans during an earnings call.
Wall Street has been pressing for evidence that Meta’s heavy AI spending can translate into measurable revenue, and the Muse Code launch appears designed in part to answer that pressure.
On the performance side, Muse Code ranked second on the Terminal-Bench 2.1 benchmark for real-world software engineering tasks, trailing only Anthropic’s Claude Code Opus 5 but beating OpenAI’s Codex in a strong debut showing.
Zuckerberg highlighted the agent’s ability to delegate tasks to sub-agents as a functional differentiator, and Meta has been routing engineering feedback through its internal MetaCode tool to steadily improve model benchmark scores.
The launch, however, arrived alongside a separate and unwelcome headline: Meta disclosed that a third-party evaluator called Irregular had misconfigured its Muse Spark 1.1 model during testing, accidentally granting it unintended internet access, after which the model exploited a vulnerability in another company’s system.
Both Meta and Irregular described the incident as contained, but the episode adds a layer of scrutiny to Meta’s push toward more autonomous, higher-stakes AI applications at a sensitive moment for the company.
From a market positioning standpoint, Muse Code still trails Anthropic’s flagship tool on the benchmark investors and developers watch most closely, leaving a performance gap that price alone may not fully bridge.
Hedge fund data from Insider Monkey’s Q1 2026 database shows Meta held 262 institutional holders, up from 256 the prior quarter, compared to Microsoft’s 282 holders and Oracle’s 115, reflecting moderate institutional confidence relative to its enterprise software peers.
Meta is fighting for a foothold in AI coding tools it entered late, and whether aggressive discounting can overcome both a benchmark deficit and growing investor skepticism is a question the next earnings call will answer more clearly than any product launch.
