What is Muse Spark 1.2?

Muse Spark 1.2 is Meta Superintelligence Labs' coding-focused update to Muse Spark 1.1, released August 5, 2026 — Meta's third model in four months. It ships alongside Muse Code, a beta terminal coding agent co-trained with the model. Standard API pricing is unchanged from 1.1 at $1.25 input / $4.25 output per 1M tokens with a 1,048,576-token context window.

Is Muse Spark 1.2 cheaper than Claude Opus 5 or GPT-5.6 Sol?

Yes, dramatically. Muse Spark 1.2 costs $1.25/$4.25 per MTok versus Claude Opus 5 at $5/$25 and GPT-5.6 Sol at $5/$30. That makes Muse Spark roughly 4x cheaper on input and 6x cheaper on output than the two frontier models, positioning it as a value coding option.

Which is best for coding: Muse Spark 1.2, Opus 5, or GPT-5.6 Sol?

Claude Opus 5 leads on raw coding benchmarks (96.0% SWE-bench Verified) and agentic/computer-use tasks. GPT-5.6 Sol is the strongest terminal-and-browsing generalist. Muse Spark 1.2 is the value pick — significantly cheaper with strong agentic coding, best when cost-per-task matters more than the last few benchmark points.

Quick Answer

Muse Spark 1.2 vs Opus 5 vs GPT-5.6 Sol (Aug 2026)

Published: August 7, 2026

The Short Answer

Muse Spark 1.2 (Meta, launched August 5, 2026) is a coding-focused, budget-priced model at $1.25/$4.25 per MTok. Claude Opus 5 ($5/$25) leads raw coding and agentic benchmarks. GPT-5.6 Sol ($5/$30) is the strongest terminal-and-browsing generalist. Pick Muse Spark for cost, Opus 5 for peak coding quality, Sol for browsing-heavy agents.

Pricing at a Glance (verified Aug 6, 2026)

Model	Input /MTok	Output /MTok	Context	Released
Muse Spark 1.2	$1.25	$4.25	1M	Aug 5, 2026
Claude Opus 5	$5	$25	1M	Jul 24, 2026
GPT-5.6 Sol	$5	$30	—	Jul 9, 2026 (GA)

A 30K-in / 5K-out task: Muse Spark ≈ $0.06, Opus 5 ≈ $0.28, Sol ≈ $0.30. Muse Spark is ~4–5x cheaper per task.

What Changed with Muse Spark 1.2

Coding-first update. Meta says it scaled up training compute on coding tasks and widened the diversity of training environments, while keeping the general-agent ability the 1.1 release sold on.
Muse Code shipped alongside it — a beta terminal coding agent co-trained with the model, Meta’s answer to Claude Code and Codex-style CLIs.
Third model in four months. After Muse Spark (post-Llama pivot) and 1.1 on July 9, this cadence puts Meta in a tie with SpaceXAI for third place among US labs on agentic knowledge work, per Artificial Analysis.

Coding & Agentic Strength

Claude Opus 5 posts 96.0% SWE-bench Verified and wins agentic/computer-use benchmarks (BrowseComp, OSWorld 2.0, AutomationBench). It’s the novel-reasoning and coding specialist.
GPT-5.6 Sol leads on DeepSWE 1.1 and long-running professional workflows (Agents’ Last Exam high of 53.6). Best for terminal + browsing generalist agents.
Muse Spark 1.2 improves code generation, complex debugging, and codebase understanding over 1.1. It won’t top Opus 5 on raw scores, but at ~5x lower cost it’s the value-per-token leader for coding agents.

Which Should You Pick?

Cost-sensitive coding agents / high volume: Muse Spark 1.2. The per-task math is decisive.
Peak coding quality, novel problems, computer use: Claude Opus 5.
Browsing + terminal generalist workflows: GPT-5.6 Sol.
Cheapest frontier-ish everyday chat: neither of these — GPT-5.6 Luna at $0.20/$1.20 undercuts all three, but trades coding depth.

Watch Outs

Muse Code is beta. Terminal agents live or die on tool-use reliability; verify on your repo before committing.
Benchmark apples-to-oranges. SWE-bench Verified is one of the few numbers reported by multiple vendors; treat single-vendor headline scores as provisional.
Intro pricing. Some frontier prices shift; re-check vendor pages before budgeting.

Verdict

If you want frontier-grade coding, Claude Opus 5 is the benchmark leader and GPT-5.6 Sol the browsing generalist. But Muse Spark 1.2 reframes the question: at $1.25/$4.25 with Muse Code as a free-standing agent, Meta is competing on cost-per-task, not leaderboards. For teams shipping high volumes of agentic coding, that’s the more important axis.

Sources

Meta AI Research — Introducing Muse Code and Muse Spark 1.2: research.meta.ai
Artificial Analysis — Muse Spark 1.2: artificialanalysis.ai
DataCamp — Claude Opus 5 vs GPT-5.6 Sol benchmarks & pricing: datacamp.com
OpenAI — GPT-5.6: openai.com