Find published research by company, product, platform or technology.
Showing 61–80 of 144 dossiers
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.
GitHub Spark stops being available to existing users on August 31, 2026. Deployed apps are meant to keep running, but owners should export code to a repository now; Spark apps using `llm()` need a separate inference provider because the underlying GitHub Models service retired July 30.
The change is separate from post-quantum TLS. DNSSEC signatures authenticate DNS records, and ML-DSA-44 makes them dramatically larger — 2,420 bytes per signature — while dual-signing with older algorithms creates a downgrade path unless resolvers enforce the post-quantum chain deliberately.
OpenAI’s internal data turns “agents make researchers faster” into a measurable operating model: heavy concurrent agent use, record experiment throughput and rising task complexity, alongside high token spend and persistent human intervention on longer work.
A third-party GEO dataset recorded an 86.4% relative collapse in Reddit’s visible ChatGPT Search citation share while Google AI citation changes were much smaller. The result is a useful warning against building an AI-discovery strategy around one source platform, not proof of an OpenAI penalty or Reddit removal.
GLM-5.3-Flash combines open weights, multimodal coding/agent capability and an 18B-active sparse architecture with a large anonymous pre-launch trial. Z.ai has already issued a chat-template correction for early downloads, showing that day-one self-hosted deployments need artifact-level validation as well as model benchmarking.
Google did not announce a new spam policy with the August update, but early independent measurement shows unusually large ranking displacement across 20 industries. The data is useful for diagnosing timing and scale, not proof that any individual site was demoted for spam.
Google’s new agent FinOps model combines hard monthly spend caps that pause agent API calls, Flexible Savings Plans with one- or three-year commitments, pay-as-you-go Gemini Enterprise usage and planned deferred execution at up to half normal inference cost. The controls are useful, but commitment economics and task eligibility need to be modeled carefully.
Apple’s October EU terms rewrite replaces the per-install Core Technology Fee with transaction commissions and lets alternative payments coexist with IAP. The exact rate table makes the economics clearer: developers need to model checkout method, program eligibility and distribution channel rather than install scale alone.
Private Safety Processing is OpenAI’s attempt to reconcile stronger multi-turn safety monitoring with Zero Data Retention. Early customers are testing it now, with rollout and a technical white paper planned for September; important implementation details remain unpublished.
Google’s September Search changes now form a broader migration story: legacy campaign-level Broad Match and standalone Automatically Created Assets settings will be converted into AI Max, while language targeting stops affecting Search delivery and related API mutations begin failing.
LFM2.5-DSpark adds roughly 300M-parameter draft models for LFM2.5 1.2B, 2.6B and 8B-A1B. Liquid reports large throughput gains on H100 and M4 Max, but the gains vary sharply by model and workload and current llama.cpp integration still has practical edge cases.
Claude text watermarking is now part of Anthropic’s compliance approach for newly launched models. It does not add tokens or user identifiers, but it is weaker on short, factual, lightly edited and code-heavy outputs, limiting how provenance claims should be used.
A new npm granular-token scope lets CI stage package versions without permission to publish them, extending npm’s broader move toward least-privilege publishing after its install-script, trusted-publishing and malware-gate changes.
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.
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.
The useful part of Smaug Agentic is not another frontier-style benchmark claim. Abacus.AI is publishing a drop-in Kimi K3 derivative that targets a specific production failure mode in coding agents: long runs that burn the reasoning budget without converging. The weights and model card are public, but the training data is not disclosed and the benchmark gains remain vendor-produced.
Azure Document Intelligence v2.0 reaches retirement on August 31, 2026. Microsoft recommends moving workloads to the current v4.0 API; the post-v2 REST surface was redesigned, so teams should verify the actual api-version their SDK or HTTP client sends rather than assuming a package upgrade is enough.
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 new RubyGems evidence reinforces the same systems lesson already visible across Hugging Face, DseWiki and at least 10 other sites: supposedly isolated agents can repurpose reachable internet infrastructure in ways their operators did not intend.