Find published research by company, product, platform or technology.
Showing 1–20 of 25 dossiers
K2 Horizon is notable less for another benchmark claim than for reproducibility: IFM is publishing model weights, architecture, training code, data or construction recipes, evaluation resources and intermediate training material instead of stopping at a final checkpoint.
The interesting signal is segmentation, not 'seat pricing is dead': vertical SaaS operators expect payments and finance to become a larger part of revenue, while horizontal platforms expect usage and hybrid software charges to lead. The survey is vendor-produced and forward-looking.
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.
Gemini 4 Argon is now available through Google's paid API and eligible subscriptions. The introductory $2 input/$10 output rates double on December 1, and its 1M-token continuation feature requires special integration.
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 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.
Demand Gen is becoming a broader acquisition system rather than only a visual campaign type: advertisers can test conversational lead capture, travel offers tied to destination context and AI-assisted horizontal/vertical video production from one campaign surface.
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.
This is a platform architecture migration rather than a user-facing feature. Pantheon says no action is required, but builders operating storage-sensitive WordPress or Drupal workloads should know when their tier moves and verify backup, restore and file-handling behavior around the change.
Two distinct access paths matter more than the benchmark headlines: Muse Wonder is available through Meta Model API and Muse Code, while Wonder Full is currently subscriber-only in Muse Code.
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 release consolidates several recurring cluster-management jobs into core APIs and controllers. HPA scale-to-zero is now default-on Beta, storage-version migration and Pod Certificates are Stable, DRA can satisfy existing extended-resource requests, and large etcd reads gain a streaming path that reduces peak memory pressure.
Muse Spark 1.3 is more than a routine model refresh: Meta is pairing stronger agent behavior with lower vendor-reported tool/token use at the same published unit price. Independent testing supports a capability gain, but max reasoning can consume substantially more reasoning tokens.
Gemini 3.8 Flash keeps 3.7 Flash’s promotional per-token rate and Flash-tier latency, but early independent analysis suggests harder reasoning can increase tokens consumed per task. A separate 3.8 Flash Cyber model is available only through Google’s Fairwind defensive-security program.
Memory-bound agents, retrieval systems and stateful services can now choose 2-, 4-, 8- and 12-CPU Render plans with much wider RAM ratios. Existing plan prices and legacy IDs stay compatible; the new choices change the cost trade-off for workloads that previously had to overbuy CPU to get enough memory.
Meta Muse Spark Open is genuinely downloadable, unlike a promised future weight release. But 42B active parameters do not make its 816B weights cheap to host, and the Muse Model License is not Apache or MIT: model-as-a-service operators over $20M in affiliated annual gross revenue need a separate agreement.
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.
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.
Cloud Run instances sit between autoscaling serverless services and a small VM. They run one individually addressable container continuously, can be stopped and restarted, and use shared CPU economics; Google’s launch example prices 1 vCPU plus 1 GiB running for 30 days at $5.70.