Cognition Slashes Devin Review 2.0 Costs by 82% per Pull Request
Cognition cut Devin Review's cost by 82% since August, added autofix loops, custom reviewers, and incremental PR passes to tackle the review bottleneck.
20 stories
Cognition cut Devin Review's cost by 82% since August, added autofix loops, custom reviewers, and incremental PR passes to tackle the review bottleneck.
Google's new video model stretches clips to 15 seconds, adds turn-by-turn editing, and drops API pricing 30% to $0.07 per second at 720p.
Perplexity's new Teammates turn its Computer agent into always-on digital coworkers that monitor work, collaborate in Slack, and run in isolated desktops.
Reflection's Beam open-weight model posts the highest harmful-request refusal rate among open peers while keeping helpfulness high, backed by a new open safety framework.
Meta's public time service now authenticates every NTP packet using RFC 8915, closing a 40-year hole in the protocol that underpins all TLS validation.
Google DeepMind's 740M-parameter multimodal embedder unifies text, code, images, video, and audio in one 768-dim space, beating models up to 12x larger.
Parallel's unified MCP server packs search, extract, decisions, and deep research into one URL, free without an API key, with admin controls for agent fleets.
Google Research built a smartwatch AI that passively detects hypertension from PPG and accelerometer signals, matching clinical cuff screening.
Prism ML squeezes a 9B vision-language model down to 2 GB with ternary weights, hybrid attention, and 262K context for on-device inference.
Spark 4.3 pushes sub-150ms stateful streaming, history-preserving CDC, and a new shuffle model into the open-source engine, with quarterly releases ahead.
Google's EnvHarness adapts existing agent training environments based on failure patterns, boosting ALFWorld held-out task success by nine percentage points
Mistral Studio adds managed agents, apps, and workflows, letting enterprises build and deploy AI systems in their own cloud or on-premises.
MiniMax releases an open-weight text-to-music model that generates full 5-minute songs in 32 kHz stereo with native ComfyUI support.
webAI post-trained IBM's Granite 4.2 8B into a reasoning model that tops MATH-500 at 98.2 and matches frontier code scores.
Switch turns Slack, Teams, Discord, Telegram and Mattermost channels into shared rooms where humans and AI agents collaborate on the same thread.
A differentiable hybrid model built on top of LPJmL beats pure physics and pure ML approaches at predicting wheat and maize yield anomalies, especially during droughts.
ComfyUI repackaged RelayVSR, a streaming video super-resolution model that upscales clips to 1080p and 4K using sparse keyframe references and a lightweight per-frame decoder.
FAH predicts 2D and 3D hand keypoints directly then fits a mesh, hitting 65 fps and 4.31 mm joint error on a single GPU.
Tiny Corp's new $50,000 tinybox red whale packs 14 AMD Radeon AI Pro R9700 cards and 448 GB of GPU memory into a dual-PSU desk chassis.
Baidu's new 280B sparse Mixture-of-Experts model activates just 14B parameters per token and ships raw pre-training weights with a 1M-token context.