Find published research by company, product, platform or technology.
Showing 181–200 of 449 dossiers
CLM-8B targets the same narrow decision layer as Jev, but with open weights, local deployment and a contrastive architecture that separates state and action representations. The headline speed and coding results are researcher-produced and need careful interpretation.
The important change is economic rather than another flagship benchmark win. OpenAI is making capable agent and coding workloads materially cheaper, with Luna approaching older Sol-class results at a tiny fraction of the task cost and GPT-6 prompt caching discounting reused input by up to 90%.
The interesting change is economic as much as benchmark-driven. Anthropic is compressing capability that previously justified its larger Fable tier into Opus pricing, while cutting Opus list prices and expanding immediate availability across the major clouds.
Branch previews are common for frontend code, but Worker Previews extends the boundary to the runtime itself. Each branch can have independent bindings, state and logs, making parallel human and agent work safer while preserving a production-like execution path.
MiMo-V2.6 is more useful than another benchmark launch because builders get both capable multimodal weights and a rare view into the reinforcement-learning machinery that produced them: code, environments, run costs and even failure notes from the training cluster.
The important change is not simply that Claude can run several agents. Projects now owns decomposition, shared context, branch isolation and progress coordination across full Claude Code sessions, while the trade-offs become usage burn, cloud-only execution and ordinary merge conflicts when parallel work overlaps.
The useful part is not the 800,000-line headline. GitHub has published unusually detailed receipts for a production-scale agent-assisted migration: roughly $120,000 of token spend, 14.5 weeks of incremental releases, dozens of regressions, extensive compatibility tests and a workload-specific jump from 7.55 to 120 session lifecycles per second.
Android Bench 2.0 moves coding-agent evaluation away from small repository fixes toward dependency upgrades, app builds, migrations and other jobs that can take a human engineer days. The results expose a much larger reliability gap than short-task benchmarks—and show that the agent harness can materially change cost and outcome.
Jev’s launch claims were interesting; Vercel’s usage data is more useful. Nearly 13% of paid AI Gateway teams tried the typed decision model in its first day, while Jev also rose to a material share of gateway requests. That does not establish retention or production success, but it is unusually fast developer uptake for a model designed to make bounded software decisions rather than generate prose.
This is not one headline vulnerability fix. Gemini CLI 0.60 is a coordinated hardening pass across the plumbing that lets extensions, sandboxes, filesystem paths and MCP authentication influence an agent’s execution environment.
The scanner itself is not the new part. The September 16 change removes the CodeQL-default-setup gate that GitHub’s July rollout originally required, making AI-assisted vulnerability detection easier to add to repositories with different code-scanning configurations.
The change separates three things that are often bundled together: the harness, the subscription that pays for it, and the sandbox that executes it. Builders can switch among supported coding agents behind one interface while reusing existing subscription access and reducing credential exposure inside agent runtimes.
The observe–test–release loop now has explicit economics: Free and Pro include 30,000 captured generations and 25 million system-initiated AI tokens per month; Pro overages start at $1.50 per 1,000 generations and $2 per million LLM Eval/Guard tokens, while ordinary telemetry is billed separately.
Astra's adoption question is no longer only model capability. Builders can now model its long-context economics and task-level efficiency, while enterprises get a more explicit control plane for computer use. The same release also raises the cyber-safety boundary: OpenAI says Astra is its first model to reach the Preparedness Framework's Critical cybersecurity capability threshold.
The material change is that model routing is no longer a single opaque optimization target. Developers can now tell Copilot whether to bias Auto toward lower cost, a middle ground or higher quality while GitHub still chooses a model prompt by prompt.
Fusion is interesting less as another routing feature than as a different agent-cost architecture: two persistent model contexts divide planning, review and execution instead of making one expensive model handle every token. The practical question for builders is shifting from token price to cost per completed task.
Google appears to have completed a talent-focused Mechanize deal: the startup still exists, but much of the team that builds coding-agent training environments and evaluations has moved into Google’s model-development work.
The material issue is not ordinary model distillation. Anthropic’s evidence suggests a customer-facing AI product may have used a rival model as an undisclosed backend while simultaneously harvesting those interactions for training, turning routing architecture into a privacy and trust boundary.
For agent and untrusted-code workloads, the useful change is not simply lower latency. Sandbox location becomes an explicit execution policy, so teams can align code execution with nearby data and avoid a resilience fallback quietly moving work outside an allowed region.
Jalapeño is working first-party silicon rather than a roadmap item, and OpenAI now says AI itself materially accelerated the design process. The distinction still matters: tape-out means the design was finalized for manufacturing; it does not mean fleet-scale production qualification or API deployment is complete.