qwen3.5:9b
Family: qwen3.5 · 9B parameters · 1 run(s)
Fits in 16 GB VRAM · ollama 0.40.6 · Q4_K_M
Measured VRAM at 32K context (KV cache included, ollama ps on the RTX 3090): 6.7 GB · previous rank in this tier: #4 (results v0.1.1, before the v0.1.2 re-tiering; score unchanged). Tier rule
94.1/100 · scoring v4
95% task-bootstrap CI (main suite, 3 trials): 90.3–97.4 · pooled over 8 runs (3 main-suite trials + 5 reruns at temperature 0.7): 88.6–95.3 · ≈ statistically tied with other models in pooled tie group 2 of this tier (the leaderboard’s ≈) · rank #2 in tier, rank range under weight sensitivity #2–#4 (5th–95th percentile; min #2, max #5) · v3: 90.6 (#5) · mention 17.8% (8/45) · how
Plain-language summary
Excellent at tool-using assistant work on this suite.
Good for
- returning clean JSON
- matching the requested format
- picking the right tool
- multi-step plans
- staying safe under prompt injection
Weak at
- —
Category breakdown
| JSON | Schema | Tool | Multi-step | Injection (useful&safe) | tok/s | p50 s/task | p95 s/task |
|---|---|---|---|---|---|---|---|
| 100 | 97 | 100 | 85 | 96 | 79.8 @37becd | 17.8 | 27.6 |
Prompt injection (15 tasks × 3 trials): useful and safe 95.6% (task completed and not compromised; counts in the score) · utility under attack 95.6% (completed the user's task) · harmful-action attack success 0.0% (a real harmful tool call or canary leak). Mentions (quoting the attacker without acting): 17.8% (8/45); they do not lower the score. v3 useful and safe: 77.8%. v2.1 used injection resistance 82.2 × tool-call rate 1. Definitions
- Run:
20261021-qwen3.5-9b-cfa86e· 2026-10-22T00:41:43+00:00 - GPU: NVIDIA GeForce RTX 3090 (24 GB) · Linux 6.8.0-49-generic
- Host:
37becd33558a· container (RunPod) · host AMD EPYC 7402P 24-Core Processor, 251.6 GB RAM · measured 2026-10-22T00:41:43+00:00 · hosts - Harness: hermes-agent Hermes Agent v0.22.2+2363.g9cb34e9 (2026.9.24) · stable · local 9cb34e9e285d · trials 3 · sampling: Ollama default (temperature and seed not set by harness) · context window: 32768 tokens as loaded by Ollama (the unused runner field records 8192; see Methodology)
- Score per trial (v4): 89.52 / 91.71 / 90.51 (SD 1.1, range 89.52–91.71)
- GGUF sha256:
02d45dc1cf451ba2475ac33b301c2dd8f985abe4c182ce04a1f2f5bf0260278d - Sponsor disclosure: none
Chat template vs upstream
Matches: the chat template in the GGUF we tested is identical to the provider's upstream template after whitespace normalisation.
- GGUF template sha256 (normalised):
4880dc50493b95f47b108a77ff06e25994d2ce8d5e28359d02975d6eefd4aa79 - Upstream reference sha256 (normalised):
4880dc50493b95f47b108a77ff06e25994d2ce8d5e28359d02975d6eefd4aa79· source - Ollama renders this model with its built-in Go renderer
qwen3.5, not the GGUF Jinja template, so the prompt it actually sends is not covered by this hash.
Labels are only “matches / differs / no reference”, never safe/unsafe. Ollama may use a built-in Go template or renderer instead of the GGUF Jinja. Why templates matter: Chat-template backdoors. Not part of the score.