What breaks when upgrading to LiteLLM v1.100.0?

LiteLLM v1.100.0 carries one marked breaking change: prompt_token_calculator has been deleted from utils. Code that imported that helper directly needs to move to LiteLLM's current token-counting path. Everything else in the release is additive or opt-in, including the new budget-rollover and fallback-authorization flags, which stay off until an operator enables them.

How do LiteLLM's shared budgets work?

LiteLLM v1.100.0 lets a budget be enforced across a model access group rather than per key, and adds a LiteLLM_BudgetWindowSpend table so spend is tracked per budget window instead of only as a running total. An opt-in rollover setting carries an overage into the next window. Admins can set a group's shared budget from the dashboard.

Which new models and providers does LiteLLM v1.100.0 reach?

LiteLLM v1.100.0 adds native Vertex AI Interactions API support, Grounding with Bing Search as a search provider, and day-0 routing for gemini-3.5-transcribe and transcribe-live, including on Vertex AI's /v1/audio/transcriptions. It also registers zai-org/GLM-5.3-Flash on Together AI, qwen-image-3.0 image generation on DashScope, and Veo 3.1 Lite metadata.

What changed for MCP users in LiteLLM v1.100.0?

The LiteLLM MCP gateway gains asymmetric RS256 signing for session tokens and RFC 7662 token introspection, so an external service can validate a gateway session. Anthropic MCP connectors can now be bulk-imported through the API or the admin UI, and a resolved OAuth token can be sent in a custom upstream header instead of the default one.

BerriAI · 2026-09-06 · major

LiteLLM v1.100.0 — Vertex AI Interactions API and shared team budgets

LiteLLM v1.100.0 adds native Vertex AI Interactions API support, Bing Search grounding, day-0 Gemini transcription models, and shared budgets across model access groups. It references 398 pull requests and deletes prompt_token_calculator.

GitHub card for BerriAI/litellm, the open-source AI gateway

The LLM gateway's newest release widens provider coverage and moves budget enforcement from single keys to shared groups.

Quick facts

Maker	BerriAI
Version	v1.100.0, released 6 September 2026
Previous stable	v1.99.0
Pull requests referenced	398
Breaking change	prompt_token_calculator deleted from utils
New providers	Vertex AI Interactions API, Bing Search grounding, gemini-3.5-transcribe
Docker image	ghcr.io/berriai/litellm:v1.100.0, cosign-signed

What is it?

Native Vertex AI Interactions API support is the headline in LiteLLM v1.100.0, alongside Grounding with Bing Search as a search provider and day-0 routing for Google's gemini-3.5-transcribe and transcribe-live models. LiteLLM is the open-source gateway that puts 100-plus model APIs behind one OpenAI-shaped interface, with cost tracking, guardrails, load balancing and logging.

How does it work?

Budgets are the structural change. A new LiteLLM_BudgetWindowSpend table records spend per budget window, so a shared budget can be enforced across a model access group instead of only per key, and an opt-in setting rolls an overage into the next window. The auto-router gains operator-defined tier sets that can be previewed and dry-run against the backend before saving. On the security side the MCP gateway adds RS256 session-token signing and RFC 7662 introspection, and /v1/rag/ingest now enforces vector-store upload controls.

Why does it matter?

Teams that put every model call through one gateway have had to approximate group-level spend limits with per-key budgets, which drift as keys are added. Enforcing a budget on the model access group, with per-window rows behind it, makes the number the finance team sees match the number the proxy enforces. The one deletion, prompt_token_calculator, is the only thing to check before upgrading.

Who is it for?

platform teams running a shared LLM proxy

Frequently asked questions

What breaks when upgrading to LiteLLM v1.100.0?: LiteLLM v1.100.0 carries one marked breaking change: prompt_token_calculator has been deleted from utils. Code that imported that helper directly needs to move to LiteLLM's current token-counting path. Everything else in the release is additive or opt-in, including the new budget-rollover and fallback-authorization flags, which stay off until an operator enables them.
How do LiteLLM's shared budgets work?: LiteLLM v1.100.0 lets a budget be enforced across a model access group rather than per key, and adds a LiteLLM_BudgetWindowSpend table so spend is tracked per budget window instead of only as a running total. An opt-in rollover setting carries an overage into the next window. Admins can set a group's shared budget from the dashboard.
Which new models and providers does LiteLLM v1.100.0 reach?: LiteLLM v1.100.0 adds native Vertex AI Interactions API support, Grounding with Bing Search as a search provider, and day-0 routing for gemini-3.5-transcribe and transcribe-live, including on Vertex AI's /v1/audio/transcriptions. It also registers zai-org/GLM-5.3-Flash on Together AI, qwen-image-3.0 image generation on DashScope, and Veo 3.1 Lite metadata.
What changed for MCP users in LiteLLM v1.100.0?: The LiteLLM MCP gateway gains asymmetric RS256 signing for session tokens and RFC 7662 token introspection, so an external service can validate a gateway session. Anthropic MCP connectors can now be bulk-imported through the API or the admin UI, and a resolved OAuth token can be sent in a custom upstream header instead of the default one.

Try it

docker pull ghcr.io/berriai/litellm:v1.100.0