# LiteLLM v1.96.0-dev.2

- Product: LiteLLM (https://whatsnew.fyi/product/litellm)
- Vendor: BerriAI
- Date: 2026-07-31
- Version: v1.96.0-dev.2
- Original notes: https://github.com/BerriAI/litellm/releases/tag/v1.96.0-dev.2
- Permalink: https://whatsnew.fyi/product/litellm/releases/v1.96.0-dev.2
- Labels: Pre-release

What's New is an index, not a publisher: every entry below links to the vendor's own release notes, which are the authoritative source. Entries are labelled where they are hand-curated sample data, pre-releases, or drawn from a secondary source such as a developer blog.

Reuse: the summaries, labels and curation here are © What's New. Quote freely with attribution and a link back; wholesale republication of the corpus is not permitted — terms: https://whatsnew.fyi/terms. The vendors' own release notes remain their publishers'.

---

- **security** — Docker images are now signed with cosign for verification
- **fixed** — Adjust gpt-5.6-terra and gpt-5.6-luna prices according to OpenAI's latest article
- **fixed** — Stamp provider on embedding cache-hit spend logs
- **fixed** — Run post_call guardrails on /v1/messages streaming via unified guardrail translation
- **fixed** — Correct Kimi K2.5/K2.6/K2.7 max output token limits for fireworks_ai
- **added** — Extend keyless gateway OAuth flow to per-server MCP URL paths
- **fixed** — Request stream usage upstream by default and strip it from client streams
- **changed** — Move the v3 limiter per-request stash off request metadata onto a ContextVar
- **fixed** — Stop compressing the turn the model must act on in guardrails/headroom
- **fixed** — Let the internal user and org forms save sub-cent budgets
- **fixed** — Serve config guardrails from list and info endpoints without a DB and make their ids stable
- **fixed** — Split mixed stream chunks by payload kind for Anthropic
- **fixed** — Give ComplexityRouter LLM classifier prior-turn context
- **added** — Mark the auto-router's own classifier calls in spend logs
- **fixed** — Bind litellm_metadata by reference in function_setup so guardrail info reaches spend logs
- **added** — Support SSE-KMS encryption params on both S3 logging paths
- **fixed** — Preserve config-defined policies across DB sync and expose them via list APIs
- **fixed** — Correct gpt-5.6 prices for openai, bedrock, and flex long context
- **fixed** — Bill the fast service tier at the priority rate
- **fixed** — Map all documented in-stream error codes to real HTTP statuses

##### Verify Docker Image Signature

All LiteLLM Docker images are signed with [cosign](https://docs.sigstore.dev/cosign/overview/). Every release is signed with the same key introduced in [commit `0112e53`](https://github.com/BerriAI/litellm/commit/0112e53046018d726492c814b3644b7d376029d0).

**Verify using the pinned commit hash (recommended):**

A commit hash is cryptographically immutable, so this is the strongest way to ensure you are using the original signing key:

```bash
cosign verify \
  --key https://raw.githubusercontent.com/BerriAI/litellm/0112e53046018d726492c814b3644b7d376029d0/cosign.pub \
  ghcr.io/berriai/litellm:v1.96.0-dev.2
```

**Verify using the release tag (convenience):**

Tags are protected in this repository and resolve to the same key. This option is easier to read but relies on tag protection rules:

```bash
cosign verify \
  --key https://raw.githubusercontent.com/BerriAI/litellm/v1.96.0-dev.2/cosign.pub \
  ghcr.io/berriai/litellm:v1.96.0-dev.2
```

Expected output:

```
The following checks were performed on each of these signatures:
  - The cosign claims were validated
  - The signatures were verified against the specified public key
```

---
##### What's Changed
* fix(pricing): adjust gpt-5.6-terra and gpt-5.6-luna prices according to OpenAI's latest article by @lihugang in https://github.com/BerriAI/litellm/pull/35258
* fix(caching): stamp provider on embedding cache-hit spend logs by @devin-ai-integration[bot] in https://github.com/BerriAI/litellm/pull/35282
* fix(proxy): run post_call guardrails on /v1/messages streaming via unified guardrail translation by @mateo-berri in https://github.com/BerriAI/litellm/pull/35260
* fix(fireworks_ai): correct Kimi K2.5/K2.6/K2.7 max output token limits by @devin-ai-integration[bot] in https://github.com/BerriAI/litellm/pull/35174
* feat(mcp)!: extend keyless gateway OAuth flow to per-server MCP URL paths by @tin-berri in https://github.com/BerriAI/litellm/pull/34856
* fix(proxy): request stream usage upstream by default and strip it from client streams by @mateo-berri in https://github.com/BerriAI/litellm/pull/35290
* refactor(rate-limits): move the v3 limiter per-request stash off request metadata onto a ContextVar by @mateo-berri in https://github.com/BerriAI/litellm/pull/35278
* test(e2e): skip the batch rate-limiter spend-row test pending LIT-5027 by @ryan-crabbe-berri in https://github.com/BerriAI/litellm/pull/35301
* fix(guardrails/headroom): stop compressing the turn the model must act on by @tin-berri in https://github.com/BerriAI/litellm/pull/35294
* fix(ui): let the internal user and org forms save sub-cent budgets by @ryan-crabbe-berri in https://github.com/BerriAI/litellm/pull/35302
* fix(guardrails): serve config guardrails from list and info endpoints without a DB and make their ids stable by @mateo-berri in https://github.com/BerriAI/litellm/pull/35259
* fix(anthropic): split mixed stream chunks by payload kind by @yucheng-berri in https://github.com/BerriAI/litellm/pull/35289
* fix: give ComplexityRouter LLM classifier prior-turn context (LIT-4981) by @tin-berri in https://github.com/BerriAI/litellm/pull/35185
* feat(spend-logs): mark the auto-router's own classifier calls by @tin-berri in https://github.com/BerriAI/litellm/pull/35300
* fix(logging): bind litellm_metadata by reference in function_setup so guardrail info reaches spend logs by @yucheng-berri in https://github.com/BerriAI/litellm/pull/35292
* feat(s3): support SSE-KMS encryption params on both S3 logging paths by @yucheng-berri in https://github.com/BerriAI/litellm/pull/35291
* fix(policy_engine): preserve config-defined policies across DB sync and expose them via list APIs by @mateo-berri in https://github.com/BerriAI/litellm/pull/35263
* test: fix order-dependent flake in passthrough guardrail call-type test by @mateo-berri in https://github.com/BerriAI/litellm/pull/35317
* fix(pricing): correct gpt-5.6 prices for openai, bedrock, and flex long context by @mubashir1osmani

_[Truncated at 4000 characters — full notes: https://github.com/BerriAI/litellm/releases/tag/v1.96.0-dev.2]_