Route the work. Not just the model.
Combining Claude, OpenAI, and Google Gemini for cost-optimized, quality-maximized enterprise workflows.
Stop picking one model. Start routing the stages.
Single-provider routing guides assign each use case to the cheapest viable model within one provider's lineup. Multi-provider pipelines go a step further: they decompose a workflow into distinct stages, then match each stage to the best-fit model from any provider based on what that stage actually requires.
Volume funnel
Run cheap classification on everything. Qualify aggressively. Deploy the premium model only on the fraction of inputs that cleared the threshold. Savings compound with volume.
Structural advantage
Use one provider because of a capability no other matches at that price tier: context window, multimodal input, or retrieval integration. Switch providers once that structural need is satisfied.
Real-time triage
Every input passes through a fast, cheap classifier first. The routing decision determines which model tier handles the response. Effective cost reflects the distribution of input complexity.
Five models. Three pipelines.
Five models appear across the three pipelines. Prices are per million tokens for input and output.
| Provider | Model | Input | Output | Context | Appears in |
|---|---|---|---|---|---|
| Google Gemini | Gemini 2.5 Flash-Lite | $0.10 | $0.40 | 1M | 01·S1, 03·S1 |
| OpenAI | GPT-4.1 Nano | $0.10 | $0.40 | 1M | 02·S2 |
| OpenAI | GPT-5.4 Nano | $0.20 | $1.25 | 400K | 01·S2, 03·S2 |
| Google Gemini | Gemini 3.1 Pro | $2.00* | $12.00* | 1M | 02·S1 |
| Anthropic | Claude Sonnet 5 | $3.00† | $15.00† | 1M | 01·S3, 02·S3, 03·S3 |
* Gemini 3.1 Pro rate applies to requests at or below 200K tokens; $4.00 / $18.00 above that threshold. † Claude Sonnet 5 standard pricing; introductory pricing of $2.00 / $10.00 applies through August 31, 2026.
Outbound prospecting at scale.
Score the entire prospect universe cheaply. Research the qualified slice. Spend the premium budget only on the top of the pyramid.
Signal detection and scoring
Google Gemini
Gemini 2.5 Flash-Lite
$0.10 / $0.40 per MTok
Every account in the prospect universe passes through this stage: potentially thousands of signals per day. The task is well-defined classification with clear output categories.
Account research and synthesis
OpenAI
GPT-5.4 Nano
$0.20 / $1.25 per MTok
Only accounts that cleared Stage 1 reach this point, typically 5 to 15% of the total prospect universe. The job is synthesis against retrieved context, not premium generation.
Personalized outreach drafting
Anthropic
Claude Sonnet 5
$3.00 / $15.00 per MTok
Claude Sonnet's writing quality is where the pipeline's premium spend is concentrated. This stage runs only for the top 1 to 5% of prospects.
Legal contract review and redlining.
Use Gemini's 1M-token window for full contract-suite ingestion in one pass, then hand off to Claude for the high-value legal output.
Full document ingestion and extraction
Google Gemini
Gemini 3.1 Pro
$2.00 / $12.00 per MTok
The 1M-token window holds most full contract suites in a single API call, eliminating chunking logic, cross-chunk reconciliation, and context bleed.
Clause classification and risk tagging
OpenAI
GPT-4.1 Nano
$0.10 / $0.40 per MTok
Extraction is complete. Stage 2 is pure classification against a risk taxonomy with high, medium, and low tags.
Redline generation and deal memo
Anthropic
Claude Sonnet 5
$3.00 / $15.00 per MTok
The premium spend goes where it matters: drafting the redlines and the deal memo a lawyer will actually read.
Customer support with escalation routing.
Every message hits the cheapest classifier first. Premium reasoning activates only on escalation, keeping effective cost per ticket close to the Nano rate.
Real-time intent and sentiment classification
Google Gemini
Gemini 2.5 Flash-Lite
$0.10 / $0.40 per MTok
A sub-200ms classification layer routes each message to the right tier before any response is generated.
Tier-1 response generation
OpenAI
GPT-5.4 Nano
$0.20 / $1.25 per MTok
Routine replies stay on the lower-cost model. Effective cost reflects the real distribution of support complexity.
Complex escalation resolution
Anthropic
Claude Sonnet 5
$3.00 / $15.00 per MTok
Premium reasoning activates for the smaller set of issues where nuance, policy interpretation, and resolution quality matter.
Per 1,000 support messages: $100 first-look classification, $140 routine replies, and $375 escalations.
Roughly 5× cost reduction, while improving routing quality for the complex cases that need it.
The plumbing under the pipeline.
Each pipeline requires an orchestration layer to pass context between stages, handle errors at each provider, and enforce stage-gating logic.
Orchestration
Each pipeline maps directly to an n8n workflow. Stage 1 is an HTTP Request node hitting the Gemini API; its output feeds a conditional branch. Stage-gating logic lives in the branch node, not in the model prompts.
Context passing
The structured output from each stage becomes part of the next stage's input prompt. Use JSON for inter-stage context, not free-text summaries, so each stage can reference fields like signal_score, clause_type, and intent_tag.
Error handling
Each provider's API can fail independently. Wrap each stage in try/catch with exponential backoff. If Stage 1 fails, do not route to Stage 2: fail the pipeline and retry from the beginning.
Provider API keys
Each provider requires its own API key and rate-limit management. Do not share rate-limit budgets across pipelines. Set per-pipeline daily spend caps and alert before limits are hit.
The takeaway
The cheapest model is rarely the right answer, and neither is the best one. The right answer is a pipeline where each stage runs on the model that fits that stage, and the premium spend concentrates where it moves the outcome.