Metacodingreasoningopen-sourcefine-tuning
Llama 3.1 405B
Llama 3.1 405B is Meta's largest open-source model, competitive with top proprietary models on benchmarks.
Arena ELO
1247
Avg Benchmark
81.2%
Context Window
128K
Speed
45 t/s
Latency P50
2100ms
Input / 1M
$0.9
Capability Profile
Category scores across Coding, Math, Reasoning, Writing, Instructions, Hard Prompts
Benchmark Scores
MMLU88.6%
HumanEval89%
MATH73.8%
GSM8K96.8%
GPQA51.1%
BBH88.1%
Pros
✓Fully open-source and free to deploy
✓No data leaves your infrastructure
✓Competitive benchmarks
Cons
✗Requires significant compute to self-host
✗No official vendor support
Pricing
Self-hosted
Free
· Full weights
· No usage limits
Together AI
Custom
· $0.90/1M tokens
· Managed hosting