Fetching from the wire…
Public story · 2026-09-08 · high
The same release also fixes a test-disappearance bug tied to Fusion manifests and blocks unsafe package install paths.
Why now: As of September 8, 2026, dbt-core 1.12.4 is the release teams running Fusion-generated manifests need most, alongside the new tracing functions.
dbt-core 1.12.4 adds two Jinja functions, otel_trace_id() and otel_span_id(), that write OpenTelemetry trace and span ids into compiled model SQL, per dbt Labs' 1.12.4 release notes. A slow or failing query now carries the id of the dbt run that fired it. Tracking a bottleneck back to a specific model no longer means lining up timestamps across two separate logging systems.
Issue #15914 covers a bug where built-in generic tests were evicted from the test registry and left undefined at compile time. That happened after a project loaded a Fusion-generated manifest, and tests just stopped running with no error to explain the gap. For any team already running Fusion, a test suite that fails without an error is a bigger problem than one that's merely slow.
The release also validates that a resolved package install path stays inside the configured packages-install-path, closing a path traversal case. The three fixes arrive together in a single patch release, not a major version bump. Anyone running Fusion-generated manifests or relying on package installs to stay sandboxed should update regardless of whether they care about tracing.
Each link below shares sources, entities, or timing with this story.
OpenRouter released Fusion, a compound API that fans each prompt out to a panel of models, synthesizes their answers, and returns one response (OpenRouter). On Perplexity's DRACO deep-research benchmark, 100 tasks across 10 domains, a Fable 5 + GPT-5.5 fusion scored 69.0% vers...
Docker-only build, no PyPI package, released September 2. OpenTelemetry v2 spans now emit gen_ai.usage.cache_creation.input_tokens and gen_ai.usage.cache_read.input_tokens alongside the cache cost already reported (release). The notes say directly that anyone computing spend f...
An Apache-2.0 OpenAI-compatible gateway at 695 stars, with per-user and per-agent spend controls, that ingests OpenTelemetry traces from existing agent workflows, builds a simulation from them, and uses that to fine-tune an open-source model you own (GitHub). Rust-native, repo...
Google's Agent Development Kit for Python listed litellm>=1.75.5 as an optional dependency. No upper bound. No pin. During the week of March 24, LiteLLM versions 1.82.7 and 1.82.8 were compromised by the TeamPCP group with a three-stage payload: credential harvesting, Kubernet...
A model addressed through a gateway as openai/gpt-5.6-luna or z-ai/glm-5.2 failed every anchored capability rule, so it got classified as a standard model, reasoningEffort was silently deleted from the request, and pricing lookup returned nothing (GitHub). Setting an effort le...
A model addressed as openai/gpt-5.6-luna failed anchored capability matching, so reasoningEffort got deleted from the request with no warning and pricing lookup returned nothing (GitHub). If you route through OpenRouter or LiteLLM, verify the effort level actually reaches the...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.