Upvote Frontier Index on Product Hunt
Frontier Index
Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)53.4 INT$20
Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)53.2 INT$20
GPT-6 Astra (max)52.7 INT$20
GPT-6 Astra (xhigh)52.4 INT$20
Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback)51.2 INT$20
GPT-6 Astra (high)50.9 INT$20
Claude Opus 5 (Adaptive Reasoning, Max Effort)50.8 INT$10
Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)49.7 INT$10
Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)53.4 INT$20
Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)53.2 INT$20
GPT-6 Astra (max)52.7 INT$20
GPT-6 Astra (xhigh)52.4 INT$20
Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback)51.2 INT$20
GPT-6 Astra (high)50.9 INT$20
Claude Opus 5 (Adaptive Reasoning, Max Effort)50.8 INT$10
Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)49.7 INT$10
Meta AI logo

western lab · United States

Meta AI

Meta's superintelligence and FAIR research groups, historically the largest publisher of open-weight frontier models through the Llama series, now paired with a heavily capitalised in-house compute program.

Founded
2013
Models tracked
22
Valuation
Meta subsidiary

est. · as of Aug 2026

Country
United States

Founders

Yann LeCun · Mark Zuckerberg

Model Benchmarks

data byArtificial Analysis
ModelReleasedIntelligenceCodingGPQAHLESciCodeTerminal-Bench HardTau2LCRIFBenchSpeedTTFT$/1M in$/1M out$/1M blended
Muse Spark 1.3 (max)2026-09-0248.175.893.5%48.7%58.8%83%290 t/s27.03s$1.25$4.25$2
Muse Spark 1.3 (xhigh)2026-09-0245.176.594.1%47.5%59.7%83%339.7 t/s21.43s$1.25$4.25$2
Muse Spark 1.2 (xhigh)2026-08-0539.672.290.4%45.5%57.4%79%0 t/s0s$1.25$4.25$2
Muse Spark 1.1 (xhigh)2026-07-0933.771.389.8%46.2%58.8%77.7%0 t/s0s$1.25$4.25$2
Muse Spark2026-04-0831.358.688.4%40.7%45.5%91.5%78%75.9%0 t/s0s$0$0$0
Muse Glimmer (high)2026-08-1017.54983.5%22%44.9%83.3%90.6 t/s0.36s$0.35$1.5$0.64
Llama 4 Maverick2025-04-051016.367.1%4.9%31.7%6.8%17.8%50%43%107.2 t/s0.59s$0.26$0.91$0.42
Llama 4 Scout2025-04-058.18.258.7%3.8%21.3%1.5%15.5%27.7%39.5%103.7 t/s0.61s$0.19$0.68$0.31
Llama 3.3 Instruct 70B2024-12-067.711.949.8%3.6%3%26.6%15.7%47.1%89.3 t/s0.62s$0.71$0.72$0.71
Llama 3.1 Instruct 405B2024-07-237.351.5%4%6.8%19%25.3%39%0 t/s0s$0$0$0

Indices on Artificial Analysis 0-100 scale · eval scores are raw accuracy · “—” means the metric is not published for that model

Release Timeline

last 48 months
  • 2026-09-02Meta AI released Muse Spark 1.3 (xhigh).Model ReleaseIntelligence index 45.1
  • 2026-08-10Meta released Muse Glimmer, a 30B open-weight (Apache 2.0) model for local agentic and coding workloads.Model Release
  • 2026-08-05Meta shipped Muse Spark 1.2, powering Muse Code, MSL's first dedicated coding agent.Model Release
  • 2026-07-09Meta Superintelligence Labs launched Muse Spark 1.1, with multi-agent orchestration and a public preview of the Meta Model API.Model Release
  • 2026-04-08Meta debuted Muse Spark, the first model from the reorganized Meta Superintelligence Labs.Model Release
  • 2025-07-26Meta named Shengjia Zhao, co-creator of ChatGPT, as chief scientist of Meta Superintelligence Labs.Milestone

Milestones written by Frontier Index from Artificial Analysis release dates, disclosed funding rounds, the official newsroom and major tech press. Linked rows open the original report.