Releasesby GlyphRay-2.58.0
Highlights Ray Serve LLM: In this release we've completed KV cache and token aware request routing, which was previewed in 2.57. Tokenization now happens in-process on the LLMRouter ingress replica, the routing decision is made there, tokens are transmitted out-of-band so the engine does not re-tokenize, KV lifecycle events are broadcast to every ingress...
Also shipped Aug 23, 2026Anyscale in August 2026 →
feldera 0.337.0
- [nexmark] Port Nexmark unit test data from Rust to SQL/Java [web-console] Fix transient test failure due to reverse scroll mechanism not aborting [SQL] Support the MODE aggregate function by desugaring into GROUP BY + COUNT + ARG_MAX [adapters] Retry Delta table opens on a transient GCS listing flake.
August 2026
Point-in-time search , query_time is now documented on both the MCP tool and the Search API : search the index as it stood at an instant, and have relative date filters resolve against that instant rather than now. Two more search parameters , max_results caps how many results come back (1–50), and snippet_max_length caps the snippet per result.
July 2026
Prompt-driven extraction , the Fetch endpoint accepts a prompt , and returns what an LLM read out of the page for that instruction instead of the whole page. Usage and overrides over MCP , every authenticated tool call now reports its cost under _meta["keenable/usage"] , and an integrator can pin the search mode with _meta["keenable/overrides"] .
August 2026
Follow each page from training crawl to assistant fetch The existing AI Traffic reports now expose the bot identity Searchable already resolves internally. Pass groupBy=botCategory or groupBy=botName on REST /traffic/overview , /traffic/bots , or /traffic/correlation , or use the same groupBy argument on the consolidated get_ai_traffic MCP tool.
More from RayEvery Ray release →
Ray-2.57.0
Highlights Ray Data: In this Ray release, we've enabled DataSourceV2 by default ( #64821 ), so read_parquet and friends use the new scan/listing infrastructure with row-group-aware chunking and predicate splitting. Hash Shuffle V2 eliminates the aggregator actor pool.
Ray-2.56.1
Highlights Ray Data : We added fixes for several to_pandas regressions introduced in 2.56: an opt-out flag ( RAY_DATA_ENABLE_ARROW_BACKED_PANDAS_CONVERSION ) for Arrow-backed conversion, an int64/ double[pyarrow] overflow crash on concatenation, and a TensorDtype.__from_arrow__ crash on empty tensor columns ( #64793 , #64794 ).
Ray-2.59.0
Highlights > * Ray Data LLM & Ray Serve LLM are GA/Stable: the LLM APIs graduate to general availability this release (\#65194), alongside an upgrade to vLLM 0.27.0 (\#65351).
Ray-2.56.0
Highlights Ray Data Stability: In this Ray release, we've added a variety of stability improvements, including running multiple datasets in a cluster, adding automatic batch size selection to CPU-based map-batches, and default logical memory configuration to prevent OOMs.
Ray-2.58.1
Fixes Ray Data - Fix out-of-scope block callbacks being registered on blocks passed through from input to output (\#66369, cherry-pick \#66404). Fixes spurious callback errors on passthrough datasets. Images - google-cloud-cli is now installed in the base-extra images (\#65757, pick \#66611). Contributors Thanks to everyone who contributed to this release!
Ray-2.57.1
Fixes Ray Data - Fix out-of-scope block callbacks being registered on blocks passed through from input to output (\#66369, cherry-pick \#66403). Fixes spurious callback errors on passthrough datasets. Images - google-cloud-cli is now installed in the base-extra images (\#65757, pick \#66610). Contributors Thanks to everyone who contributed to this release!
Ray-2.60.0
Ray 2.60.0 released shuffle v2 as default for joins and aggregations, updated Parquet task sizing from row-group metadata, added rolling update surge controls in Ray Serve, and patched security vulnerabilities.
Ray-2.59.1
Ray 2.59.1 released security fixes updating bundled certifi, pip, and virtualenv dependencies following the eMudhra intermediate CA compromise, along with updating aiohttp to 3.14.4.