qwen3.8:27b
Family: qwen3.8 · 27B parameters · 1 run(s)
Fits in 24 GB VRAM · ollama 0.40.6 · Q4_K_M
Measured VRAM at 32K context (KV cache included, ollama ps on the RTX 3090): 17 GB · previous rank in this tier: #5 (results v0.1.1, before the v0.1.2 re-tiering; score unchanged). Tier rule
98.7/100 · scoring v4
95% task-bootstrap CI (main suite, 3 trials): 97.4–99.7 · pooled over 8 runs (3 main-suite trials + 5 reruns at temperature 0.7): 90.0–95.0 · ≈ 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–#3 (5th–95th percentile; min #2, max #3) · v3: 92.0 (#5) · mention 33.3% (15/45) · how
† The published score (98.7, main-suite mean of 3 trials) lies above the pooled 8-run interval (90.0–95.0): the 5 reruns at temperature 0.7 scored lower (rerun mean 92.91). Score and rank stay those of the main suite. Why
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 | 99 | 100 | 96 | 100 | 61.0 @b7e2c7 | 15.2 | 25.4 |
Prompt injection (15 tasks × 3 trials): useful and safe 100.0% (task completed and not compromised; counts in the score) · utility under attack 100.0% (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): 33.3% (15/45); they do not lower the score. v3 useful and safe: 66.7%. v2.1 used injection resistance 66.7 × tool-call rate 1. Definitions
- Run:
20261023-qwen3.8-27b-25675d· 2026-10-23T03:28:09+00:00 - GPU: NVIDIA GeForce RTX 3090 (24 GB) · Linux 6.8.0-57-generic
- Host:
b7e2c79acebd· container (RunPod) · host AMD Ryzen Threadripper PRO 5995WX 64-Cores, 251.5 GB RAM · measured 2026-10-23T03:28:09+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): 93.81 / 94.54 / 87.62 (SD 3.8, range 87.62–94.54)
- GGUF sha256:
f5f1dd8920d417aac2718b0bda3403da274301efdd6760b4f0f4b864ff2ad57d - 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):
094b7f7d201492fcc2d2410519b0d836957d998bc51bae5007d0be14bd858949 - Upstream reference sha256 (normalised):
094b7f7d201492fcc2d2410519b0d836957d998bc51bae5007d0be14bd858949· source - Ollama renders this model with its built-in Go renderer
qwen3.8, 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.