Cognition, the corporate behind the Devin coding agent, has launched SWE-2, its most succesful coding mannequin thus far. SWE-2 is post-trained with reinforcement studying from Kimi K3, Moonshot AI’s 2.8T-parameter open mannequin. Cognition reviews a rating of fifty.0% on FrontierCode 1.1 Main, inside 1 level of Fable 5.1 at 64% decrease price. It can also…
The Fly Language Model (FLM) is a public chatbot that {couples} the full retained MaleCNS v1.0 fruit fly connectome to a frozen LiquidAI LFM2.5-1.2B-Instruct spine. The developer who created the FLM calls it the world’s first Fly Language Model, constructed on an structure known as GPF (Generative Pre-trained Fly). It doesn’t use the GPF label,…
An agent harness is the code round a mannequin: execution loop, instruments, context, state, restoration, and verification. Per the Terminal-Bench 2.1 leaderboard, GPT-5 solves 35.2% of duties inside Terminus 2 however 49.6% inside Codex CLI with similar weights. Most benchmarks preserve that harness fastened. HarnessDev proposed by staff of researchers from ByteDance Seed, Singapore University…
Anthropic has printed a new plugin evals workflow for Claude Code. The claude plugin eval command runs a plugin towards life like prompts, grades what Claude produced, and compares the outcome with a run the place the plugin will not be loaded. It solutions 3 questions plugin builders couldn’t beforehand measure: does the ability set…
Let me ask you one thing. In your group proper now, what proportion of your staff use AI as a lot as you do? When I requested that query at a current AI Accelerator Institute event, fewer than 10% of fingers went up. And these had been chief AI officers. People whose literal job is…
NVIDIA is utilizing Palantir Foundry and cuOpt to automate its {hardware} supply chain allocation selections throughout world manufacturing websites. The firm measures operational supply from wafer-out to first token. This window splits into time-to-rack (the transit from fab output to an assembled information centre system) and time-to-token (which covers energy, cooling, networking, and day-one software…
This article is sponsored by CDD Vault and was written, edited, and printed in alignment with our Emerj sponsored content guidelines. Learn extra about our thought management and content material creation providers on our Emerj Media Services page. Drug discovery is one among the slowest, costliest processes in enterprise R&D. Developing a single FDA-approved remedy…
Cohere has launched North Small Translate, an open-weight machine translation mannequin from Cohere and Cohere Labs. It is a sparse Mixture-of-Experts (MoE) mannequin with 218B complete and 25B energetic parameters. It covers 50 languages, from Albanian to Vietnamese. On Cohere’s WMT26 analysis, it scores 83.6 averaged throughout all languages. Cohere says that beats DeepL and…
Sakana AI has launched Fugu Max and Fugu Ultra v2, 2 new fashions in its Sakana Fugu household. Fugu isn’t a single basis mannequin. It is a discovered orchestrator that routes work throughout a pool of different fashions behind 1 API. The new launch tunes that structure for 2 missions. Fugu Max targets one of…
Training an LLM to name instruments reliably requires datasets that pair person queries with right tool-use chains. Producing that information at scale has been gradual and costly. A workforce of researchers from Google, the University of Tokyo, RIKEN AIP, and Tohoku University introduce ToolGrad. The analysis work inverts the standard pipeline: construct a verified device…