Meta open-sources Llama 4.1 with improved agentic reasoning and 256K context

Read the original article →

What happened

Meta released Llama 4.1 as open weights on August 5 in 8B, 70B and 405B parameter sizes. The update emphasizes multi-step agentic reasoning, 256K context windows and improved function-calling reliability compared with Llama 4.0.

Meta said cumulative downloads of Llama family models passed 700 million since the Llama 3 launch. The release includes updated responsible-use guidelines and new benchmark suites focused on agent and tool-use tasks.

Why it matters

Open-weight momentum depends on models that are good enough for agents, not just chat. Llama 4.1's agent focus and long context make it a stronger self-hosted option for teams that cannot send data to closed APIs.

Meta's download numbers also show how widely open models have spread across hobbyists, startups and enterprises.

MintedBrain take

Match model size to your hardware before committing. The 8B and 70B variants are practical for many teams; the 405B tier is only relevant if you already run large GPU clusters. Test function-calling on your real tools, not Meta's benchmarks alone.

References

This article was originally published at Hugging Face. For the full piece, read the original article.

Discussion

  • Loading…

← Back to News