← Leaderboard

qwythos-9b-claude-mythos-5-1m

empero-ai/qwythos-9b-claude-mythos-5-1m empero-ai/qwythos-9b-claude-mythos-5-1m on HuggingFace ↗
9B vLLM General

Throughput

14 7 0 12.6 4k 12.4 50k 12.3 100k tok/s

Decode tok/s after a fixed context fill (PBM) — same agent-style workload as bench v2.

Recipe

Profile
empero-ai-qwythos-9b-claude-mythos-5-1m-eugr
Engine
vLLM
Context
781k · fp8 KV
Served as
qwythos-9b-claude-mythos-5-1m

Run on your Spark

single node shell
spark inference up empero-ai-qwythos-9b-claude-mythos-5-1m-eugr

Why we run it

Golden fleet target — auto-scaffolded from recipe empero-ai-qwythos-9b-claude-mythos-5-1m-eugr.

Bench notes

PBM 4k @ 12.6 tok/s — perfbench-metrics — profile=empero-ai-qwythos-9b-claude-mythos-5-1m-eugr

Measurement history

Context ladder

Older bench-v2 / golden cells at each benched context window.

Context KV Throughput
@ 256k fp8 12.2t/s
@ 320k fp8 12.0t/s
@ 400k fp8 12.0t/s
@ 500k fp8 11.8t/s
@ 600k fp8 11.6t/s
@ 781k peak golden fp8 12.4t/s
@ 1M fp8 10.8t/s

Benchmark runs

Recorded inference benchmark sessions for this model's profile (single run).

Date Profile Method Avg Session t/s Range Fill Tool
2026-06-28latest empero-ai-qwythos-9b-claude-mythos-5-1m-eugr bench-agent-v2v2.0 12.4t/s ~50,000 fail
golden 781k/fp8 @ 12.4 tok/s — fill~50000 — bench-agent-v2 — tool_ok=False
Benchmarked 2026-07-10
SparkBench · GB10 · single node