Claude Code × LiteLLM compatibility matrix
This table is regenerated daily by an automated populator that runs the
Claude Code CLI against
the newest final LiteLLM release (the latest bare vX.Y.Z tag) across each
supported provider, with Haiku 4.5, Sonnet 4.6, and Opus 4.7 in parallel. A
cell goes green only if all three model tiers pass.
| Feature | Anthropic | Bedrock (Invoke) | Bedrock (Converse) | Vertex AI | Azure (Foundry) | openai | azure_openai | bedrock_mantle | vertex_ai_gpt |
|---|---|---|---|---|---|---|---|---|---|
| Basic messaging (non-streaming) | ✅ | ✅ | ✅ | ✅ | ✅ | — | ✅ | — | n/a |
| Basic messaging (streaming) | ✅ | ✅ | ✅ | ✅ | ✅ | — | ✅ | — | n/a |
| Tool use | ✅ | ✅ | ✅ | ✅ | ✅ | — | ✅ | — | n/a |
| Prompt caching (5m TTL) | ✅ | ✅ | ✅ | ✅ | ✅ | — | — | — | — |
| Vision | ✅ | ✅ | ✅ | ✅ | ✅ | — | — | — | — |
| Thinking | ✅ | ✅ | ✅ | ✅ | ✅ | — | — | — | — |
| Tool use (streaming / fine-grained) | ✅ | ✅ | ✅ | ✅ | ✅ | — | ✅ | — | n/a |
| Extended thinking + tool use | ✅ | ✅ | ✅ | ✅ | ✅ | — | — | — | — |
| PDF document input | ✅ | ✅ | ✅ | ✅ | ✅ | — | — | — | — |
| Prompt caching (1h TTL) | ✅ | ✅ | ✅ | ✅ | ✅ | — | — | — | — |
| Web search (server tool) | ✅ | ✅ | ✅ | ✅ | ✅ | — | — | — | — |
| Structured outputs | ✅ | ✅ | ✅ | ✅ | ✅ | — | — | — | — |
| count_tokens endpoint | ✅ | ✅ | ✅ | — | ✅ | — | — | — | — |
| Tool search (MCP discovery) | ✅ | ✅ | ✅ | — | — | — | — | — | — |
| Native API passthrough | ✅ | ✅ | n/a | ✅ | — | — | — | — | — |
| Long context (1M) | — | — | — | — | — | — | — | — | — |
Legend
| Glyph | Meaning |
|---|---|
| ✅ | All three model tiers pass for this (feature, provider) cell. |
| ❌ | At least one model tier failed. Hover for the upstream error. |
| — | No test ran for this combination. |
| n/a | Not applicable (e.g. provider doesn't expose this feature). Hover for the reason. |
Source
The matrix JSON lives at
src/data/compatibility-matrix.json.
The populator is in
tests/e2e/claude_code/cron_vm/
on the main repo.