Meta Launches Muse Spark 1.3: Frontier Performance at a Fraction of the Price

Episode 1 · Meta launches Muse Spark 1.3, matching Fable 5 at a fraction of the price (2026-09-03, 53 posts)

Meta officially released Muse Spark 1.3 on 09-03, calling it its biggest capability leap yet in coding and agentic tasks, with the pitch of "frontier performance at a price too cheap to meter." The model is live on Muse Code and the Meta Model API, with the max reasoning tier coming after safety testing is complete; Zuckerberg confirmed the release and teased that the next model and an open-weights version of Muse Spark are coming soon.

Confirmed

• Priced at $0.10 per million input tokens and $0.20 per million output tokens — an ultra-low price roughly one-eighth of Claude Fable 5's pricing (per Alexandr Wang's comparison)
• This is Meta's fourth iteration in five months and its third consecutive upgrade from 1.1 to 1.3 in two months
• On the Artificial Analysis Intelligent Index, the xhigh version scored 61-62, a continued jump from 57 for 1.2 and 53 for 1.1, tying Claude Fable 5 and surpassing GPT-5.6 Sol, Grok 4.6, Kimi K3, and Gemini (per m3, m5); Reddit users also shared screenshots claiming it beats GPT-5.6 Sol
• On Artificial Analysis's Coding Agent Index, the xhigh version sits on the Pareto frontier at $1.72 per task; the max version can't be computed as Meta hasn't published its pricing
• Official improvement data vs 1.2: about 20% fewer tool calls, about 25% fewer tokens consumed, stronger long-horizon tasks and visual coding; supports longer task chains across multiple workflows in a single thread, proactively asks clarifying questions, flags stuck states, and confirms before critical actions
• Third-party benchmarks (per armandruiz): coding capability surpasses GPT-5.6 and Opus 5; ChrisGPT's comparison with Gemini 3.8 Flash claims long-context scores exceed GPT-5.6 (data cited from Google DeepMind, not independently verified)
• aimlapi ran a hands-on 1.3 vs 1.2 comparison using a "Viking figurine" single-file HTML generation task (per alexandrwang)
• Muse Code offers one-line installation: curl -fsSL <url> | bash

Unconfirmed

• m10's benchmark data cited from Google DeepMind is not independently verified; details of m2's leaderboard screenshots haven't been published

Why it matters

• Meta is delivering near-Claude Fable 5 Intelligent Index performance at about one-eighth the price, while occupying the cost-performance Pareto frontier for Coding Agents — direct pressure on frontier-model pricing
• The four-iterations-in-five-months pace shows Meta's momentum in catching up on model capability, and the teased open-weights version further widens its potential impact

• Meta Releases Muse Spark 1.3 With Major Gains in Agentic and Coding Performance — alexandr_wang · 2026-09-03
• Muse Spark 1.3 Cuts Tool Calls ~20% and Tokens ~25%, Meta Says — alexandr_wang · 2026-09-03
• Meta's Muse Spark 1.3 Cuts Tool Calls ~20% and Tokens ~25%, Says Alexandr Wang — alexandr_wang · 2026-09-03
• Meta Releases Muse Spark 1.3, Its Strongest Model for Agentic and Coding Tasks — shengjia_zhao · 2026-09-03
• Zuckerberg Confirms Muse Spark 1.3 Launch, Teases Open-Weights Releases — ren_hongyu · 2026-09-03
• Muse Spark 1.3 rolls out with 'biggest jump' on coding and agentic work, open weights teased — shuchaobi · 2026-09-03
• Meta releases Muse Spark 1.3 with ~20% fewer tool calls and ~25% fewer tokens — AIatMeta · 2026-09-03
• Muse Spark 1.3 rolls out in Muse Code and Meta Model API, max reasoning pending — AIatMeta · 2026-09-03
• Meta teases Muse Spark open weights and bigger models ahead — AIatMeta · 2026-09-03
• Meta ships muse spark 1.3 with stronger agentic and coding skills at near-zero cost — rohanpaul_ai · 2026-09-03
• Muse launches Spark 1.3, a proactive agentic model update, third release in three days — bowenc0221 · 2026-09-03
• Meta ships Muse Spark 1.3, its third release in two months, targeting coding and agents — ren_hongyu · 2026-09-03
• Muse Spark ships 1.1 to 1.3 in two months, pushing coding and agent skills — EdwardSun0909 · 2026-09-03
• Meta releases Muse Spark 1.3 with upgraded agentic and coding performance — ns123abc · 2026-09-03
• Meta Ships Muse Spark 1.3, Third Release in Two Months, Pushing Coding and Agents — billyuchenlin · 2026-09-03
• Meta's Muse Spark 1.3 tops Gemini 3.8 Flash on most overlapping benchmarks, crushes long-context MRCR — ChrisGPT · 2026-09-03
• Muse Spark 1.3 Ships With 'Almost Too Cheap to Meter' Pricing; Open Weights Next — bowenc0221 · 2026-09-03
• Meta officially releases Muse Spark 1.3 with major agentic and coding upgrades — alexandr_wang · 2026-09-03
• Meta's Muse Spark 1.3 hits 61-62 on AA Intelligence Index, cheapest per-task at its level — ArtificialAnlys · 2026-09-03
• Muse Spark 1.3 (xhigh) is cheapest per-task at its intelligence level at $0.55 — ArtificialAnlys · 2026-09-03

33 more related posts →

Episode 2 · Meta Releases Muse Spark 1.3 as Its Strongest Model (2026-09-03, 2 posts)

Meta has released Muse Spark 1.3, its strongest model yet, with Alexandr Wang claiming it beats GPT-5.6 Sol on coding. Polymarket traders remain skeptical, giving Meta only a 12% chance of holding the top AI model by year-end.

• Meta releases Muse Spark 1.3, claims it beats OpenAI's GPT-5.6 Sol at coding — Polymarket · 2026-09-03
• Polymarket Gives Meta Just 12% Odds to Lead AI Race by Year-End Despite Muse Spark 1.3 — Polymarket · 2026-09-03

Episode 3 · Meta's Muse Spark 1.3 Runs 2-Hour Self-Improvement Loop for Under $1 (2026-09-03, 4 posts)

A user benchmark showed Meta's Muse Spark 1.3 completed a 2-hour self-improvement loop with 60+ agent calls for under $1, while Fable 5.1 cost $13 for a task done in a minute. Scale AI CEO Alexandr Wang amplified the result, and developers praised its speed and low price.

• Alexandr Wang Touts Meta Muse Spark 1.3: 1 Minute vs Fable 5.1's 70 Minutes and $13 — alexandr_wang · 2026-09-03
• Meta Muse Spark ran 60+ agent loops in 2 hours for under $1, wowing Scale CEO — alexandr_wang · 2026-09-03
• Meta's Muse Spark ran 20 self-improvement loops with 60+ agent runs in 2 hours for under $1 — TinfoilTricorn · 2026-09-03
• Early Users Praise Meta's Muse Spark 1.3: Fast, Frontier-Level and 'Too Cheap' — Scobleizer · 2026-09-03

Episode 4 · Meta Chief AI Officer Praises muse spark 1.3 Sub-Agent Demo (2026-09-03, 2 posts)

Meta's Chief AI Officer Alexandr Wang shared a developer's hands-on test of muse spark 1.3, calling it "pretty cool." The developer deployed parallel sub-agents that completed the task in about an hour.

• Meta's chief AI officer touts muse spark 1.3 as dev fans out sub-agents to finish task in an hour — alexandr_wang · 2026-09-03
• Meta's chief AI officer touts muse spark 1.3 as dev fans out sub-agents to finish task in an hour — alexandr_wang · 2026-09-03