Tag
Cognition launches SWE-1.7, a highly capable AI model for agentic software engineering that achieves frontier-level performance at reduced cost, with improvements in RL training, multi-cluster infrastructure, data curation, and self-compaction for long tasks.
PACE introduces a proxy method to estimate LLM agentic capabilities using cheaper single-turn tasks, reducing the cost and time of full agentic evaluations.
GLM 5.2 tops PostTrainBench while being 5x cheaper than Opus 4.8 and 11x cheaper than Fable 5, making personalized AI economically viable for companies and countries.
A new, low-cost Chinese AI model is rapidly advancing and competing with top US AI companies like Anthropic and OpenAI.
Google DeepMind launches Nano Banana 2 Lite, its fastest and most cost-efficient image model, and Gemini Omni Flash for video generation and conversational editing, now available to developers via Google AI Studio and APIs.
Google DeepMind released Nano Banana 2 Lite, a fast and cheap image generation model that balances quality and speed, available across the Google ecosystem.
agent-data is a Python API tool that provides structured web data for AI agents like OpenClaw, claiming to be 70% cheaper and more reliable than browser automation.
Open-weights models have caught up with proprietary ones, with GLM 5.2 achieving near Opus-level scores in browser agent tasks at low cost. Other models like Minimax M3 and Kimi k2.7 also show notable improvements.
Azalia Mirhoseini highlights DeLM, a decentralized language model approach where agents communicate via shared state, achieving ~10% improvement on SWE-bench Verified with Gemini-3 Flash at less than half the cost.
Introduces Behavior Forecasters (BFs) that take reasoning trajectories as input and achieve more accurate forecasts than frontier models at a fraction of the cost.
Irys introduces Stateful Swarms, an open-source paradigm for AI agents using structured blackboard memory to improve performance and reduce cost. On Harvey AI's Legal Agent Benchmark, it achieved an 83.74% criteria pass rate at $1.30 per task, compared to the state-of-the-art 10.4% at $50.90.
This paper introduces an adaptive on-the-fly multifidelity machine learning algorithm for quantum chemistry that autonomously determines training data composition across fidelities, reducing data generation costs by up to 30x compared to single-fidelity methods and up to 5x compared to standard multifidelity methods.
Jensen Huang hints at more Nemotron model releases, highlighting open-source frontier intelligence and cost efficiency enabled by NVFP4 training.
Mimo V2.5 offers performance comparable to Claude Opus 4.5 at a fraction of the cost, making it a highly cost-effective AI model for agentic tasks.
A developer used Codex 5.5 as an orchestrator and Deepseek v4 pro as an executor to generate a 240M token fine-tuning dataset, burning 359M tokens at a cost of only $78.
TRACER is a tool that replaces up to 90% of LLM classification calls with lightweight traditional ML by learning from LLM traces, reducing cost while maintaining accuracy.
Perceptron Inc. released its flagship video analysis model Mk1, claiming 80-90% lower cost than competitors while achieving strong performance on spatial and video reasoning benchmarks.
Interfaze introduces a hybrid AI model architecture combining CNN/DNN specialization with transformer capabilities, achieving superior accuracy on deterministic tasks like OCR and translation while maintaining cost efficiency at scale.
Google releases Gemini 2.5 Flash-Lite as stable and generally available, the fastest and lowest-cost model in the Gemini 2.5 family at $0.10 input/$0.40 output per 1M tokens, featuring native reasoning capabilities and full feature parity with native tools.
Google announces general availability of Gemini 2.5 Flash and Pro models, and introduces Gemini 2.5 Flash-Lite in preview—a new cost-efficient and fastest variant optimized for high-volume, latency-sensitive tasks.