Benchmarked against: Anthropic — Models overview
Tools: list_models, call_model, search_web
Decision: Engine Selection Guide (Captain decision, 2026-02-27)
SuperPortia operates a multi-engine fleet — six AI providers with different cost profiles, capabilities, and specializations. Unlike single-model systems, SuperPortia matches the right engine to each task, optimizing for the minimum cost that gets the job done right.
The engine fleet
Primary engine: Claude
| Model | ID | Best for |
|---|
| Opus 4.6 | claude-opus-4-6 | Architecture, decisions, complex reasoning, code operations |
| Sonnet 4.6 | claude-sonnet-4-6 | Standard coding, execution, analysis |
| Haiku 4.5 | claude-haiku-4-5 | Fast responses, high-volume tasks |
Claude is the only engine with full tool access (file operations, code editing, shell commands). All other engines are text-only — they can analyze and respond but cannot modify files or run commands.
Low-cost engines
| Engine | Provider | Default Model | Cost | Strengths |
|---|
| Groq | Groq | Llama 3.3 70B | Free | Fast inference, research, simple analysis |
| Gemini | Google | 2.5 Flash | ~$0.014/search | Authoritative search with citations, embeddings |
| DeepSeek | DeepSeek | R1 / V3 | Cents/request | Deep reasoning, analysis tasks |
| Mistral | Mistral | Latest | Cents/request | European alternative, general tasks |
| Zhipu | Zhipu AI | GLM-5 | Cents/request | Chinese NLP, agent tool-calling |
Search-augmented engines
Two engines support built-in web search:
| Engine | Mode | Use case |
|---|
Groq Search (groq-search) | Groq Compound with web search | Free intel/research tasks |
Gemini Search (gemini-search) | Gemini + Google Search Grounding | Authoritative research with citation links |
Engine capabilities matrix
| Capability | Claude | Groq | Gemini | DeepSeek | Mistral | Zhipu |
|---|
| Text generation | Yes | Yes | Yes | Yes | Yes | Yes |
| Web search | Via tools | Compound | Grounding | No | No | No |
| File operations | Yes | No | No | No | No | No |
| Code execution | Yes | No | No | No | No | No |
| Tool calling | Full MCP | XML format | JSON | Limited | Limited | JSON |
| Embeddings | No | No | Yes (embedding-001) | No | No | No |
| Chinese NLP | Good | Limited | Good | Good | Limited | Excellent |
Architecture
Checking available engines
Returns available providers, their default models, and whether API keys are configured.
Known limitations
| Engine | Limitation | Workaround |
|---|
| Groq | XML tool calling format — breaks nested agent mode | Use flat mode (automatic) |
| Groq | Token explosion on ingest_text calls | Ingest handled at orchestrator layer |
| Groq | Hallucinated version numbers for tech info | Verify with Pre-Flight Check |
| DeepSeek | Limited tool calling support | Use flat mode |
| Mistral | Limited tool calling support | Use flat mode |
| All non-Claude | No file/shell operations | Only Claude engine for WOs requiring file ops |
Related pages