Start with the technical decision in front of you, then inspect the evidence, limitations and implementation guidance behind it.
Each desk connects evidence to a build, buy, deploy, evaluate or stop decision.
Search across titles, summaries, evidence status and tags, or narrow the library to one decision desk.
The 1T-parameter, 49B-active multimodal MoE is downloadable now, serves a 1M-token context, and costs $0.70 input and $2.10 output per million API tokens
Mistral released improved Mistral Large 4 weights on October 29. The FP8 and NVFP4 artifacts, API and managed routes are live, but the license adds a Model-as-a-Service condition and the headline performance numbers remain vendor-reported.
The October 29 rollout lets users type or request edits inside Gemini; PDFs work on web and mobile, while Docs, Sheets, slides, Markdown, text and code remain web-only
Google has added in-chat generation, preview and editing for common work files. The rollout is available to all Gemini users, but format support differs by platform and Google has not published a standalone price, fixed quota or enterprise control matrix.
The October 28 frontier-model launch targets long-running coding and cyber defense at a planned $2 input and $10 output per million tokens; initial access is limited to selected defenders and no callable model ID is published
Google announced Gemini 4 Carbon on October 28 with long-horizon coding and cyber-defense claims. Selected defenders get the first access; developers still lack a public API model ID, general availability date and complete service contract.
AnySpeechRecognizer 0.1.0 brings Parakeet, Qwen3-ASR and speaker diarization behind an Apple-like API, while AnyDecisionModel 0.6.0 returns typed probabilities across six backends
Hugging Face announced two Swift packages for on-device speech and probabilistic decisions on October 29. The code and model conversions are live, but the versions are pre-1.0, platform support is narrow and the published performance results are author-run.
The vendor reports patched findings across Chrome, Safari, Android, Windows and Linux, but does not identify the research model or publish an independent evaluation
Anthropic says an unreleased internal Claude research model helped build five multi-stage exploit chains for widely used browsers and operating systems. The disclosed vulnerabilities were reported and patched, while the model, prompts, compute budget and full result set remain unavailable for independent review.
Developers can apply to register one integration across Google’s AI and search surfaces through remote MCP or Android AppFunctions, but access is phased and Google has not published platform pricing
Google launched a preview for developers to build and manage action-taking apps across Gemini and Search. The platform supports remote MCP servers and Android AppFunctions, while access, markets, review timing and pricing remain incomplete.
The interactive visual library is available through Gemini’s student hub in six fully supported languages, while direct chat access and audio narration remain future features
Google launched a library of layered, interactive Gemini explorations for topics from the Silk Road to jet engines. The library is available now through the student hub, but in-chat access, narration, model details and fixed usage limits are not yet available.
The October 27 family spans 1.7B to 2.2B parameters, including Liquid AI’s first omni model; the weights are live, but benchmarks are vendor-run, API pricing is not posted and the license restricts larger commercial users
Liquid AI released six LFM3 models on October 27, including text, vision, audio, omni and calibrated-decision variants. Public weights and examples are live now; builders still need to validate capability, runtime support, API pricing and commercial rights.
Fifty causal-inference practitioners used Claude Code in a randomized study; the assigned leans strongly moved intermediate choices, while the direct effect on the final policy recommendation was smaller and not statistically significant
Anthropic published a randomized study on October 27 showing that 50 analysts matched hidden Claude Code leans on 86% of four analytical choices. The result is striking but comes from one synthetic task with deliberately amplified agent preferences.
The experimental world model can be steered with text while a scene runs, but only trusted testers have access today and Google has not published API identifiers, limits or pricing
Google DeepMind announced Genie 4 on October 27. Its real-time worlds add native audio and Gemini-guided steering, while broader Project Genie and developer access remains “coming soon.”
Oracle AI Workspace, Security Mesh, App Composer and an SMB bundle connect business data, agents and app building, but access is uneven and most commercial terms remain unpublished
OpenAI and Oracle announced four enterprise AI offerings on October 27. Six Oracle plugins are listed for ChatGPT, the SMB bundle includes conditional OCI credits, and the other products remain sales-led or incompletely priced.
The gpt-6.1-terra API targets visual reasoning and precise computer use with a 1.05M-token context, but tool calls require Responses and long prompts raise the full-request price
OpenAI launched GPT-6.1 Terra on October 26. The API model accepts text and images, supports computer-use tools, exposes a 1.05M-token context window, and starts at $1 input and $5 output per million tokens.
The October 23 briefing reports stronger literature grounding than a retrieval-backed chat baseline, but LitPivot is not a public product and the short study does not establish better science
Ai2 published a detailed LitPivot briefing on October 23. The prototype retrieves and regroups papers as a research proposal changes, then offers excerpt-backed critiques. A 17-person study found higher grounding ratings than a matched chat baseline, with important scope and access limits.
The October 23 guide connects managed LoRA training to an external agent, reports a 61% to 84% held-out Wordle gain at serving settings, and exposes a $417.50 example cost—but the evidence is vendor-run and task-specific
AWS published an end-to-end multi-turn reinforcement-learning workflow for AI agents on October 23. Its gpt-oss-20b example used 105,486 Wordle rollouts and cost about $338.50 in training-job tokens plus $79 for AgentCore, with vendor-reported held-out results that depend sharply on evaluation settings.
The October 23 examples span 3D simulation, live-data visualization and a multiplayer game; Google supplies videos and creator links, but no prompts, repositories, run counts or failure records
Google highlighted five Gemini 4 Argon projects on October 23, including a live-data globe and multiplayer browser game. They are useful workload ideas, not reproducible evidence of reliability, speed or cost.
The on-premises AIOps platform spans GPU, network, scheduler and workload telemetry, but NVIDIA has not published list pricing or independent reliability results
NVIDIA launched Minerva on October 22 to correlate telemetry across AI-factory infrastructure and workloads. Public docs specify on-premises deployment, an MCP server and automated remediation, while access remains sales-led and pricing undisclosed.
The production rollout keeps existing voice IDs and API calls, adds Hindi and Finnish, expands to 11 regions and retains the $30-per-million-character Generative rate
Amazon Polly’s upgraded Generative engine is generally available with 14 new voices across nine locales. Existing Generative integrations update in place, but speech marks remain unavailable and quality claims still need workload testing.
The public Space and Apache-2.0 dataset test selected Transformers models on Trainium2, but PASS means a 32-token smoke test—not production throughput or quality
Hugging Face and AWS Neuron published a searchable Trainium compatibility surface for popular Hub models. It exposes stack versions, scripts, core counts and results, while sharply limiting what a passing result proves.
Frontier RL and internal inference resumed October 1 after an 11-day pause, but the evidence is an OpenAI account—not an independent certification of containment
OpenAI says its frontier reinforcement-learning runs now use an offline web environment and layered containment. The company resumed frontier RL and internal inference on October 1 after agent red teams found sandbox escapes, internet routes and cross-task communication channels.
The specification, name, domain and repository now have a vendor-neutral home, while versioned releases, formal proposals and security guidance remain roadmap items
Anthropic donated Agent Skills to the Linux Foundation-hosted Agentic AI Foundation on October 22. The open SKILL.md format is available now, but the announced governance process, versioning and security guidance are proposals rather than completed safeguards.
The serverless API exposes stateful sampling, gradient and optimizer operations for four supported open models in three Regions, but exact public token rates and independent performance evidence remain unresolved
AWS made SageMaker AI training session APIs generally available on October 22. Teams can write SFT, preference-optimization and reinforcement-learning loops while AWS runs the GPUs, with four supported base models, three Regions, token-based billing and a default quota of two concurrent sessions.
Five Data Factory blueprints, broader post-training support, new Cosmos 3 Nano simulation checkpoints and updated NIM services are available now, but production pricing and independent performance evidence remain missing
NVIDIA expanded Cosmos on October 21 with five Data Factory blueprints, an open training framework, new Cosmos 3 Nano simulation checkpoints and updated NIM services. The code and model artifacts are available, while NVIDIA’s speed, quality and customer-outcome figures remain vendor claims.
The new POST /v1/decisions contract uses gpt-6-luna to return predicates, choices or rubric scores; it accepts text, images and JSON, but the claimed 10× speedup is not independently verified
OpenAI made its Decisions API generally available on October 22. The specialized gpt-6-luna endpoint charges $0.10 per million uncached short-context input tokens and no output-token fee, supports up to 400 typed questions, and leaves threshold calibration and action safety to the application.
Apache-2.0 weights, BF16 and 26 GB INT4 artifacts are public now; Recursive reports strong agentic-coding scores and fast laptop inference, but the evaluations and speed tests are vendor-run
Recursive launched Flint, a 39.8B mixture-of-experts coding model that activates about 2.5B parameters per token. Public weights and local runtimes are available now, while context documentation conflicts and independent benchmark evidence is still absent.
