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
- 18
- Valuation
- Meta subsidiary
- Country
- United States
est. · as of Aug 2026
Founders
Yann LeCun · Mark Zuckerberg
Model Benchmarks
updated just nowartificial analysis ↗| Model | Released | Intelligence | Coding | GPQA | HLE | SciCode | Terminal-Bench Hard | Tau2 | LCR | IFBench | Speed | TTFT | $/1M in | $/1M out | $/1M blended |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Muse Spark 1.1 (xhigh) | 2026-07-09 | 50.6 | 71.3 | 89.8% | 45.1% | 58.2% | — | — | 63.3% | — | 213.3 t/s | 2.23s | $1.25 | $4.25 | $2 |
| Muse Spark | 2026-04-08 | 43.1 | 58.6 | 88.4% | 39.9% | 51.5% | 45.5% | 91.5% | 69.7% | 75.9% | 0 t/s | 0s | $0 | $0 | $0 |
| Llama 4 Maverick | 2025-04-05 | 14.3 | 16.3 | 67.1% | 4.8% | 33.1% | 6.8% | 17.8% | 46% | 43% | 107.4 t/s | 0.66s | $0.27 | $0.91 | $0.43 |
| Llama 4 Scout | 2025-04-05 | 10 | 8.2 | 58.7% | 4.3% | 17% | 1.5% | 15.5% | 25.8% | 39.5% | 120.7 t/s | 0.58s | $0.18 | $0.66 | $0.3 |
| Llama 3.3 Instruct 70B | 2024-12-06 | 9.4 | 11.9 | 49.8% | 4% | 26% | 3% | 26.6% | 15% | 47.1% | 83.9 t/s | 0.65s | $0.6 | $0.72 | $0.63 |
| Llama 3.1 Instruct 405B | 2024-07-23 | 8.5 | — | 51.5% | 4.2% | 29.9% | 6.8% | 19% | 24.3% | 39% | 0 t/s | 0s | $2.5 | $10 | $4.38 |
| Llama 3.1 Instruct 8B | 2024-07-23 | 7.6 | 5.4 | 25.9% | 5.1% | 13.2% | 0.8% | 16.4% | 15.7% | 28.6% | 0 t/s | 0s | $0.07 | $0.09 | $0.08 |
| Llama 3.1 Instruct 70B | 2024-07-23 | 6.8 | — | 40.9% | 4.6% | 26.7% | 3% | 15.2% | 6.3% | 34.4% | 0 t/s | 0s | $0.56 | $0.56 | $0.56 |
| Llama 3.2 Instruct 90B (Vision) | 2024-09-25 | 6.2 | — | 43.2% | 4.9% | 24% | — | — | — | — | 0 t/s | 0s | $0 | $0 | $0 |
| Llama 2 Chat 7B | 2023-07-18 | 4.3 | — | 22.7% | 5.8% | 0% | — | — | — | — | 0 t/s | 0s | $0.05 | $0.25 | $0.1 |
Indices on Artificial Analysis 0-100 scale · eval scores are raw accuracy · “—” means the metric is not published for that model
Newsroom
source ↗- 2026-07-09Introducing Muse Spark 1.1Model Launch
- 2026-07-27Reimagining Independence: How Meta’s AI Models Are Helping the University of Pittsburgh Transform Assistive RoboticsRelease
- 2026-07-21How Meta’s AI Models Are Powering the First Wave of Genesis Mission ProjectsRelease
- 2026-07-07Introducing Muse Image and Muse VideoModel Launch
- 2026-06-29From Brain Waves to Words: Brain2Qwerty Offers a New Path to Communication Without SurgeryRelease
Partners & Investors
disclosedFunding rounds
- Corporate R&DMeta-funded
Meta Platforms · since 2013
- Strategic stake$14.3B
Meta → Scale AI · Jun 2025
Partners
- NVIDIAChips
- Microsoft Azure (Llama)Cloud
- AWS / Google Cloud (Llama)Cloud
- Hugging FaceDistribution
- Scale AIResearch
Publicly announced rounds and partnerships only · as of Aug 2026