gemma4:12b
Family: gemma4 · 12B 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): not measured (no valid ollama ps size for this run; tier kept as published). Tier rule
98.3/100 · scoring v4
95% task-bootstrap CI (main suite, 3 trials): 96.9–99.5 · pooled over 8 runs (3 main-suite trials + 5 reruns at temperature 0.7): 97.0–99.1 · rank #1 in tier, rank range under weight sensitivity #1–#1 (5th–95th percentile; min #1, max #1) · v3: 98.3 (#1) · mention 0.0% (0/45) · 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 | 95 | 100 | 97 | 100 | 130.3 @37becd | 19.6 | 32.5 |
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): 0.0% (0/45); they do not lower the score. v3 useful and safe: 100.0%. v2.1 used injection resistance 100.0 × tool-call rate 1. Definitions
- Run:
20261022-gemma4-12b-b67c82· 2026-10-22T06:41:17+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 2026-10-22T06:41:17+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): 96.83 / 98.2 / 100 (SD 1.59, range 96.83–100)
- GGUF sha256:
bb722270d54346adc198f213851dbe0207e87c3ecf2a0aff4d92262726215391 - 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):
ac8703caf03928968acdcb598d83a5f1027c366588e60554d07d0c4490b1174a - Upstream reference sha256 (normalised):
ac8703caf03928968acdcb598d83a5f1027c366588e60554d07d0c4490b1174a· source - Ollama renders this model with its built-in Go renderer
gemma4, 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.