An engineer with 18 months of experience building AI agent systems across multiple teams shares hard-won lessons about shipping agents to production, highlighting the gap between benchmark performance…
Ornith released the Ornith-1.5 family of open-source coding models on August 20, 2026, claiming its 397B mixture-of-experts flagship scores 86.1 on Terminal-Bench 2.1 and rivals Claude Opus 4.8, thoug…
Microsoft introduced MAI-Code-1-Flash, a 5B active-parameter coding model designed for efficiency and integrated with GitHub Copilot and Visual Studio Code. The company reports up to 60% fewer tokens …
Cognition CEO Scott Wu argues that traditional AI benchmarks are losing their meaning as models saturate every test, with the company's AI coding agent Devin approaching near-perfect scores on SWE-Ben…
3code, a new coding agent, used 75% fewer tokens than OpenCode on a 10-task subset of SWE-Bench, succeeding on a task where OpenCode failed. The benchmark used the same model, GLM 5.2, across both age…
Researchers propose Process-Scorer Guided Adaptive Tree Rollout (PATR), a quality-aware rollout framework for multi-turn reinforcement learning (RL) that uses process feedback to selectively branch fr…
The Ornith-1.0 family of open-source coding models was released under the MIT license, achieving state-of-the-art scores on agentic coding benchmarks including 82.4 on SWE-Bench Verified and 77.5 on T…
XAI's Grok 4.5 scored 83% on Terminal Bench, outperforming many costlier models in multi-step coding tasks, while Anthropic's Claude Opus 4.8 excels in complex reasoning and high-stakes professional w…
Ornith-1.0, a family of open-source LLMs specialized for agentic coding, achieves state-of-the-art performance on multiple coding benchmarks including SWE-Bench and Terminal-Bench. The models range fr…
Google Labs researchers introduced a new benchmark for evaluating proactive AI coding agents, arguing that agents should be graded on their 'insight policy'—the ability to identify what matters and wh…
Microsoft expanded its in-house coding model MAI-Code-1-Flash to all GitHub Copilot surfaces on June 18, coinciding with the launch of usage-based billing. The 137B-parameter sparse MoE model costs up…
A developer model selection guide for June 2026 compares Claude Opus 4.8, Gemini 3.5 Pro, and GPT-5.6, finding that each excels on specific dimensions: Opus 4.8 leads on coding tasks within 200K token…
Reve and Ideogram both launched new image-generation models on June 2, 2026, with each company emphasizing advances in layout control through improved labeling and code. Ideogram 4.0 is now ranked as …
According to the article, a team building an agent for the SWE-Bench coding evaluation found that simple tools like `grep` and `find` were more effective than complex embedding-based retrieval, as the…