Vercel Python Queues SDK, TypeScript 5.5 type guards

Bun 1.4 lands on Vercel Functions

Bun 1.4 runtime now available on Vercel Functions with Zig-to-Rust rewrite, requiring explicit opt-in via vercel.json bunVersion config.

Reduces cold-start friction for Bun workloads on Vercel and unlocks WebSocket support via Bun.serve(). Breaking changes require deliberate testing before upgrading existing deployments.

Replaces previous Bun runtime on Vercel. Requires setting bunVersion: "1.4.x" in vercel.json and reviewing breaking changes before upgrading. Ready now for new projects; existing deployments need validation.

• “Bun 1.4 release, a rewrite of Bun from Zig to Rust, with over 2,900 issues resolved”
• “Runs on Fluid compute with Active CPU pricing”
• “Supports WebSockets and Bun.serve()”
• “Upgrading is an explicit opt-in because Bun 1.4 has several breaking changes”

v0 connects apps to 100+ third-party services

Vercel Connect lets v0-generated apps authenticate with Slack, Google, GitHub, and others via short-lived tokens managed by Vercel or custom credentials—skip long-lived secrets entirely.

Eliminates OAuth setup friction and secret rotation overhead when building multi-service integrations. Connectors are team-scoped and reusable across apps, reducing per-app auth configuration to a single prompt.

Replaces manual OAuth flows and credential management for v0 users. Requires Vercel account and v0 access; some providers (Slack, GitHub) need no provider-side setup, others accept custom credentials. Worth trying now if you're already in v0's workflow—authentication UX is materially simplified.

• “Apps and agents built in v0 can now securely connect to 100+ third-party services”
• “Connectors belong to your team, so you set one up once and reuse it across apps”
• “each connection mints a short-lived token, so there's no long-lived secret to store or rotate”

Vercel Functions now emit custom metrics natively

Use `metric()` from `@vercel/functions` to record application data alongside built-in observability, queryable via CLI and dashboards.

Eliminates external metrics pipeline setup for Vercel-deployed functions. Custom attributes enable filtering and grouping without schema migration, reducing observability tool sprawl.

Replaces ad-hoc logging for performance/business metrics. Requires `@vercel/functions` package and Pro+Observability tier. Ready now—single function call, no infrastructure changes.

• “You can now emit your own metrics directly from your Vercel Functions and analyze them alongside Vercel's built-in observability data”
• “metric("database.query_ms", 120, { table: "my-table", db: "my-database" })”
• “Custom metrics are billed as observability events and available on Pro and Enterprise with Observability Plus”
• “Vercel automatically adds useful attributes like the source deployment and function region”

Rust 1.98.0 stabilizes algebraic float operations

Algebraic floating-point methods (add, sub, mul, div, rem) enable compiler reordering for vectorization and performance, trading determinism for speed without undefined behavior.

Developers can opt into faster float math by calling algebraic_add/sub/mul/div/rem instead of standard operators, unlocking loop vectorization and partial-sum parallelization without manual SIMD code. Buffered integer formatting (format_into) now matches itoa performance, eliminating a common external dependency.

Replace itoa dependency with format_into for integer formatting immediately—it's production-ready. Algebraic float methods are an opt-in tier-2 optimization for hot paths; adopt in numerically-heavy code only if you've profiled and accepted non-deterministic results. ManuallyDrop<Box<T>> fix is a correctness guarantee, not an action item unless you hit this edge case.

• “The exact set of optimizations is not specified, but may be similar to the kind of optimization you would see with the -ffast-math option in other languages.”
• “These methods are non-deterministic, since the compiler is free to choose different optimizations, but they never cause undefined behavior.”
• “This method also bypasses much of the dynamic dispatch that you would get with buffered write! formatting, which can be a boon to performance.”
• “the itoa-benchmark repo now shows that format_into performs similarly to itoa itself, so this could serve as a standard replacement for that dependency”

Data Point

LitReview Arena benchmarks literature review generation

Agentic LLMs like Sonar Deep Research outperform base models by 60% on lit reviews, but still lose to humans 77% of the time; expert-calibrated evaluator (LitJudge) now available to measure synthesis quality.

If you're building or evaluating AI-assisted research tools, existing LLM-as-judge methods misalign with expert preferences (ρ=0.467). LitJudge fixes this gap (ρ=0.78), letting you optimize for criteria that actually matter: structure, synthesis, research suggestions.

This replaces generic ROUGE/overlap metrics for lit review evaluation. Requires domain expertise to validate outputs; code and data are public. Worth adopting now if you're building research assistants, but don't expect to replace humans—current best systems win only 23% of head-to-head matches on overall utility.

• “agentic LLMs such as Sonar Deep Research substantially outperform base language models by over 60%”
• “even the strongest current systems win only 23.0% of decisive matches against human drafts on overall utility”
• “existing LLM-as-a-judge methods are substantially misaligned with human experts (Spearman's rho=0.467)”
• “LitJudge, which improves alignment to Spearman's rho=0.78, comparable to inter-expert consistency”