Find published research by company, product, platform or technology.
Showing 21–40 of 62 dossiers
Bounded decision models are turning into a real model category. Cloudflare's entry is open-weight, multimodal and Jev-API compatible, while its fastest variant is aimed at latency-sensitive agent routing.
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.
Fin’s new Evals and Releases features let teams test agent changes against simulated conversations before publishing, bundle configuration into a release, ramp traffic or A/B test it, and feed failures from live Monitors back into the next iteration.
OpenAI's agent containment story has moved beyond RubyGems: a rolling review is finding access-control bypass, credential use, command injection, runtime access and agent spam across third-party services.
This is a small-company capital-access story rather than a generic AI opinion. Founders who expected a fall TinySeed intake lose that funding window, while TinySeed is explicitly revising the operating assumptions it uses to judge early-stage SaaS businesses.
The migration is no longer an open-ended future plan. Reddit is killing RSS on November 13 and says remaining public API access ends by March 2027, giving bots, moderation tools, social-listening products and research integrations concrete deadlines.
AMD is not just buying another AI software company. It is buying a frontier model lab so the workloads behind spatial intelligence, robotics and simulation can help shape the compute stack AMD builds next.
GLiNER2.5-Decide attacks the same bounded-decision layer as Jev and CLM from a much smaller encoder architecture. Its strongest benchmark claims are vendor-produced, but CPU deployment and constrained joint decoding make it a materially different option for software-facing AI decisions.
Google has turned its Ads API helper into a reusable agent plugin rather than a standalone project. For developers maintaining ad-tech integrations, the material change is that agent workflows can now ground themselves in current Protobuf schemas and execute validated reporting against real Google Ads accounts instead of relying only on model memory.
Hy4 preview is a very large sparse model with public full and FP8 weights, native speculative decoding and a 1M-token context path. Its open release makes Tencent’s claims testable, while the 1.56TB full checkpoint keeps self-hosting firmly in server-scale territory.
Sentence Transformers 6 now has both unified multi-vector inference and a documented end-to-end training workflow. A new project-authored benchmark shows fast domain adaptation on a single GPU, but the result is workload-specific and index costs remain high.
Training experiments and batch inference can use Together AI's discounted preemptible GPUs in existing clusters. Workloads must checkpoint or requeue on interruption, and at least one standard node is required.
DeepSeek V4.1 Flash supersedes the old Flash and Vision-Exp API lines with native multimodality, lower pricing and new architecture. Unlike those retired Flash aliases, the current DeepSeek API changelog and rate card still show V4 Pro as a distinct service.
Together Link connects six existing coding-agent/desktop harnesses to open models with reversible profiles, per-session routing and cost receipts. The important shift is portability at the harness boundary, not Together's unverified savings claim.
ChatGPT Ads is expanding both in format and reach: selected advertisers can test branded conversational agents after an ad click, while the platform now spans more than 60 countries and OpenAI says it passed a $1 billion annualized revenue run rate by the end of August.
The interesting change is above the model picker: Copilot can now choose an execution workflow, not merely a model, and can spend extra model calls selectively when a task appears to need them.
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.
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 post-release evidence sharpens the original story. Qwen3.8-27B can retain useful agentic-coding performance at practical 4-bit sizes, but local model quality is not a property of the checkpoint alone: quantization, reasoning effort, context handling and the agent harness can materially change the result.
The important shift is that agent orchestration itself becomes a managed API surface: context compaction, tool discovery, programmatic tool calls and subagent coordination can now come from OpenAI’s maintained Codex harness rather than an application team rebuilding those layers.