not ready Current free model
Google: Gemma 4 31B (free)
google/gemma-4-31b-it:free
Denne side er genereret fra Hermes LLM knowledge registry. Den viser status, research, test-evidens og næste testplan. Free-status er midlertidig metadata, ikke sidens formål.
Capabilities
include_reasoningmax_tokensmin_preasoningresponse_formatseedstoptemperaturetool_choicetoolstop_atop_ktop_p
Input modalities: image, text, video
Recommended use cases
- Reasoning/planning tasks with explicit constraints and step checks
- Structured extraction or validated JSON outputs, with parser validation
- Tool-using agent tasks with evidence-grounded final answers
- Vision or multimodal analysis where image input is needed
Skills and prompt patterns
- Mini-skill: use tools before claims, cite source/tool IDs, do not invent results, no production authority
- Reasoning guardrail: final answer must separate evidence, assumptions, and recommendation
Best practices
- Use external source evidence before final recommendation; keep local eval evidence authoritative for behavior.
- Start with low-temperature deterministic tests before creative tasks
- Log provider, returned model, latency, status, score, prompt/skill version, and redacted errors
- Persist every result immediately, including bad outcomes
- Do not test capabilities that catalog/provider metadata says are unsupported
- Prefer native tool/function calling over brittle strict JSON when tool support exists
- Validate structured outputs with an external parser; do not trust raw format claims
Test evidence
Status counts
{"429": 40, "200": 130, "404": 2, "503": 4}
Provider counts
{"unknown": 46, "OpenInference": 107, "Google AI Studio": 23}
Bad signals
{"429_cooldown_or_provider_rate_limit": 40, "404_not_available": 2, "empty_output": 4}
Recent records
| Lane | Scenario | Status | Provider | Score | Signal |
|---|
| scenario_battery | t1_smoke_danish_exact | 200 | Google AI Studio | 100.0 | |
| optimal_tools_skills | optimal_policy_dry_run | 200 | OpenInference | 100.0 | |
| optimal_tools_skills | optimal_research_synthesis | 200 | OpenInference | 100.0 | |
| optimal_tools_skills | optimal_file_diagnosis | 200 | OpenInference | 100.0 | |
| scenario_battery | t1_smoke_danish_exact | 429 | | 0.0 | 429_cooldown_or_provider_rate_limit |
| scenario_battery | t1_smoke_danish_exact | 200 | Google AI Studio | 100.0 | |
| scenario_battery | t1_smoke_danish_exact | 429 | | 0.0 | 429_cooldown_or_provider_rate_limit |
| scenario_battery | t1_smoke_danish_exact | 429 | | 0.0 | 429_cooldown_or_provider_rate_limit |
Next test plan
researchsmokemodel_specific_optimaltools_skillsworkflow_langgraph_mockvision_multimodalfinal_recommendation_pagevalidator_loopstructured_output_validationreasoning_planning
No skipped lanes recorded.
Research sources
External source enrichment present: yes
Recommendation status
not ready: Do not publish as a final recommended model yet. Keep the info page as evidence and run the listed follow-up tests.
Recommended roles
monitoring_only_until_retest
Risks and caveats
- hard: 429_cooldown_or_provider_rate_limit (40)
- hard: 404_not_available (2)
- soft: empty_output (4)
- policy: free_status_is_transient (1)