Everyday AI staff
For high-volume work that should feel immediate and economical.
- Email & meeting work
- Document extraction & summaries
- Routine operational analysis
SERVAI MODELS & INTELLIGENCE · CURATED AUGUST 2026
You don't choose from hundreds of AI models. ServAI chooses the right ones.
Tell ServAI what matters: speed, privacy, depth or cost. ServAI selects a small set of models that are actually useful for AI employees and business work, configures the runtime, and routes each task to the right intelligence.
NO MODEL HUNTNO CODING-MODEL CATALOGUENO FORCED PROVIDERHUMAN-CONTROLLED CHANGES
WORK ENTERS SERVAIROUTING POLICY ACTIVE
Meeting summary
Supplier analysis
Executive Mission
ServAIINTELLIGENCE ROUTER
FASTLocal / efficient
SMARTERProfessional work
DEEPESTPremium reasoning
01 · CHOOSE THE PRIORITY, NOT THE MODEL
Three simple intelligence levels.
ServAI keeps the technical choices behind the platform. Customers choose the business priority; ServAI handles model fit, runtime, configuration and task routing.
Professional depth
For harder finance, operations, procurement, risk and cross-functional reasoning.
- Comparisons & diagnosis
- Business reasoning
- Multimodal documents
High-value Missions
For difficult multi-step decisions where stronger reasoning justifies more compute.
- Executive synthesis
- Scenario & challenge
- Complex Mission work
ServAI is model-flexible but model-opinionated. We do not expose every model that exists. Coding-first and unrelated specialist models stay out of the customer catalogue unless they directly improve ServAI workforce work.
02 · SERVAI RECOMMENDED · AUGUST 2026
A small current set. Chosen for work.
These are current open-weight candidates ServAI considers relevant to business agents. Final availability depends on ServAI compatibility validation and the customer's hardware. ServAI does not publish borrowed tokens-per-second results as if they were measured on ServAI devices.
OPENAI · OPEN WEIGHTS
FAST REASONINGgpt-oss-20b
A strong candidate for responsive local reasoning and specialized agent work where low latency matters.
Agent tasksReasoningRoutine professional work
LOCAL / SELF-HOSTEDSERVAI COMPATIBILITY GATE
GOOGLE · OPEN WEIGHTS
BUSINESS MULTIMODALGemma 4 26B A4B
A multimodal candidate for documents, visual business inputs and general professional workloads with an efficient active-parameter design.
DocumentsCharts & imagesGeneral business work
LOCAL / SELF-HOSTEDHARDWARE DEPENDENT
MISTRAL AI · OPEN WEIGHTS
AGENTIC + MULTIMODALMistral Small 4
A current hybrid candidate for multimodal professional work, agentic tasks and complex reasoning.
Agent workflowsVisual inputsComplex reasoning
LOCAL / SELF-HOSTEDSERVAI EVALUATION REQUIRED
OPENAI · OPEN WEIGHTS
MAXIMUM OPEN DEPTHgpt-oss-120b
The high-depth open-weight option for difficult reasoning when the deployment hardware can support it.
Deep reasoningExecutive synthesisHigh-value Missions
HIGH-END LOCAL / SERVERNOT FOR EVERY DEVICE
QWEN · OPEN WEIGHTS
EFFICIENT GENERAL REASONINGQwen3-30B-A3B
An efficient mixture-of-experts candidate for high-throughput general reasoning and agent work.
General reasoningAgent workHigh-throughput tasks
LOCAL / SELF-HOSTEDSERVAI EVALUATION REQUIRED
03 · TOKEN & COST DISCIPLINE
Pay premium-model rates for premium-model work — not everything.
ServAI can reduce unnecessary paid-model usage by selecting relevant context, reusing structured evidence, handling eligible tasks locally and routing premium calls only where stronger reasoning is justified.
TYPICAL UNORCHESTRATED FLOW
Everything goes to the expensive model.
Long history, repeated context and routine work can all consume paid inference.
Full historyREPEATED
Large contextEVERY CALL
Premium modelALL WORK
→
SERVAI-OPTIMIZED FLOW
Only the context and intelligence the task needs.
ServAI prepares the task before a paid model receives it.
Relevant contextSELECTED
Local work firstWHEN ELIGIBLE
Premium reasoningWHEN NEEDED
Context selectionSend what matters, not everything available.
Evidence reuseReuse structured verified facts instead of re-sending whole sources.
Model routingKeep eligible routine work away from premium APIs.
NO FIXED SAVINGS CLAIM: actual token and cost reduction depends on workload, selected provider, caching support, data shape and routing policy. ServAI should measure customer-specific savings before publishing percentages.
04 · THREE SOURCES OF INTELLIGENCE
Native. Local. Connected. Never locked in.
ServAI's workforce layer stays separate from model choice, so a customer can keep sensitive work local, use a commercial provider for selected tasks, and adopt OUTAI when its ServAI release gates are complete.
Run supported open models.
ServAI downloads, configures and runs compatible models on supported customer infrastructure. The customer does not need to manage runtimes, quantization or inference settings.
BEST FOR · privacy · offline-capable deployments · predictable local usage
Bring a provider you already use.
Connect a supported commercial provider such as OpenAI, Anthropic or Google, or an approved compatible endpoint. ServAI can reserve those calls for workloads you authorize.
BEST FOR · frontier capability · burst capacity · existing provider agreements
OUTAI is coming to the intelligence layer.
OUTAI is being developed as ServAI-native intelligence for AI employees, Missions and business work. Individual OUTAI models remain marked Coming Soon until they pass their release, evaluation and deployment gates.
STATUS · COMING TO SERVAI · NOT FORCED
05 · OUTAI · NATIVE, NOT FORCED
The model layer ServAI can be built around.
OUTAI is intended to become the native ServAI option for supported workloads, while keeping open-model and paid-provider choice available. A production customer's model is not silently changed just because the recommended catalogue evolves.
DESIGNED AROUNDAI employees + Missions
BUSINESS CONTEXTRoles + Industry work + evidence
DEPLOYMENTPer-model hardware compatibility
RELEASE POLICYComing Soon until cleared
06 · SERVAI CONFIGURES THE TECHNICAL LAYER
The customer should never need to become an inference engineer.
Behind one simple choice, ServAI handles the technical details required to make the model useful to the workforce.
Model compatibilityMatch model class to hardware and workload.
Runtime setupInstall and configure the supported inference runtime.
Memory & quantizationChoose a compatible representation where required.
Context policyControl how much evidence and history is sent.
Health checksKnow whether the selected model is actually available.
Task routingAssign routine and deep work to the appropriate intelligence.
PRODUCTION CONTROL
Recommended catalogue updatesAllowed
Silent customer model switchNo
Paid API usePolicy-controlled
Local-only modeWhere compatible
Performance claimOnly after ServAI benchmark
INTELLIGENCE WITHOUT THE MODEL HUNT
You choose the outcome. ServAI chooses the intelligence.
Start with your work, your privacy preference and your infrastructure. ServAI handles the model layer behind the workforce.