qwen3:14b
Family: qwen3 · 14B 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): 14 GB · previous rank in this tier: #3 (results v0.1.1, before the v0.1.2 re-tiering; score unchanged). Tier rule
92.0/100 · scoring v4
95% task-bootstrap CI (main suite, 3 trials): 89.0–94.6 · pooled over 8 runs (3 main-suite trials + 5 reruns at temperature 0.7): 87.4–93.8 · ≈ statistically tied with other models in pooled tie group 2 of this tier (the leaderboard’s ≈) · rank #5 in tier, rank range under weight sensitivity #4–#5 (5th–95th percentile; min #2, max #5) · v3: 90.6 (#4) · mention 6.7% (3/45) · how
Plain-language summary
Excellent at tool-using assistant work on this suite.
Good for
- returning clean JSON
- 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 | 76 | 100 | 87 | 98 | 75.0 @37becd | 22.4 | 41.8 |
Prompt injection (15 tasks × 3 trials): useful and safe 97.8% (task completed and not compromised; counts in the score) · utility under attack 97.8% (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): 6.7% (3/45); they do not lower the score. v3 useful and safe: 91.1%. v2.1 used injection resistance 93.3 × tool-call rate 1. Definitions
- Run:
20261022-qwen3-14b-c137a5· 2026-10-22T02:46:30+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 2026-10-22T02:46:30+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): 90.18 / 95.03 / 86.66 (SD 4.2, range 86.66–95.03)
- GGUF sha256:
a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e - Sponsor disclosure: none
Chat template vs upstream
Differs: the GGUF template is not identical to the provider's current upstream template after whitespace normalisation. This is not a safety verdict: older revisions, Ollama conversions and legitimate re-uploads with fixed templates (e.g. Unsloth) often differ.
- GGUF template sha256 (normalised):
ccd61ecea6aa8406e336da83f2f71e500777096cdb0391303614e7331e6087b0 - Upstream reference sha256 (normalised):
5922e70169a13b0ee856ed984180068cf3b8369fee76cce63c8ca2d37007f0b4· source
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.