Public KG record
Together Compiler
Recorded overview
A continual compilation loop using coding agents to rewrite and optimize model f
An evidence-backed entity from the project knowledge graph; detailed tool metadata is still being added.
Recorded details
- Benchmarks: 1,111 vs 633 P90 TPS per user on MiniMax M3.1 Flash (B200, batch size 1)
- Benchmarks: 417 vs 329 P90 TPS per user on GLM 5.3 (B200, batch size 1)
- Benchmarks: 450 vs 423 P90 TPS per user on Kimi K3 (GB300, batch size 1)
- Benchmarks: up to 1.7x over Hazy Research megakernels, Mirage and CUDA-graph baselines on Llama 1B at batch size 1
- Deployment: cloud
- Maturity: limited beta for Together customers
- Performance: 252x higher agent experiment throughput than an engine-restart baseline
- Performance: found a solution 7.2% faster with two tiers
Evidence coverage
The KG currently records 10 known information dimensions and has 9 undocumented dimensions, linked to 1 public source articles. These counts describe record coverage, not model performance or verified fit.
Related public records
- Together AI — develops
- Persistent Engine — based on
- GLM-5.3 — evaluated on
- Kimi 3 — evaluated on
- Llama 1B — evaluated on