Benchmarks

Performance, side by side.

Independent BenchLM leaderboard scores for frontier and open-weight models — so you can pick the right model for the job, not just the biggest name.

Frontier models

Official OpenAI & Anthropic models, provided by Tara Cloud as a reseller.

The same official APIs, on the same terms — one API key, unified JPY billing, and a single invoice through Tara Cloud.

Official APIs resold by Tara Cloud, with unified JPY billing.

Performance per need

85% of the leader's score

Qwen3.8 Max delivers 85% of the leader's benchmark performance with a 1M context window — a strong fit for high-volume production workloads.

78% of the leader's score

GLM-5.3-Flash reaches 78% of the leader's score with a compact footprint — built for throughput-sensitive pipelines.

Open-weight models

Open-weight models hosted on Tara Cloud's own GPU infrastructure in Japan — every prompt and completion stays in-country.

All data stays in Japan.

Benchmark data: BenchLM (benchlm.ai), snapshot of September 8, 2026. Scores are indicative and updated periodically. benchlm.ai

Tell us about your workloads and we'll size the right mix of open-weight and frontier models — usually with a proposal within two business days.