Welcome Beam is a 501 billion parameter open model you can own, customize, and scale.
Reflection's Beam is a 501 Billion Apache-2.0 open LLM with 1 Million token context for coding, agents, and GPU deployment.
Reflection's Beam is a 501 Billion Apache-2.0 open LLM with 1 Million token context for coding, agents, and GPU deployment.
Microsoft Research Focus #98 reviews AI translation, a solved geometry conjecture, coding-agent costs, and a 104-language tokenizer.
Software Engineering CoWork tests coding agents in simulated workplaces; eleven models pass tests but rarely complete tasks, delivery gaps remain.
Cursor used local Noto instead of a font store, producing working but wrong output. AI coding needs reuse discipline, not more generated code.
SparseRead previews files, lets agents request relevant sections, and cuts token use up to 92.9% across tasks, making long-log debugging cheaper.
GPT-5.1 is now available through the OpenAI API, bringing faster adaptive reasoning, extended prompt caching, and stronger coding performance. Two new tools, apply_patch and shell, enable coding agent…
Video-Coder trains code agents to build 3D draft scenes, improving video structure and motion with open data, models, and gains of 7 and 24 points.
Frontier3DBench tests coding agents on 145 3D tasks; text generation works, but reference reconstruction and coordination lag.
A visual quadrant maps structural variance and execution outputs to assess LLM coding agent reliability without ground truth.
In 2026, coding agents moved from novelty to infrastructure as Claude Opus 4.5, GPT-5.1, OpenClaw, and security issues reshaped developer workflows.
Caper gates coding agent commands by pre-executing them in sandboxed traces and reviewing file, network and secret effects, lowering harmful outcomes.
Co-Design Bench evaluates coding agents on 273 multimodal app editing tasks, using acceptance criteria and evidence to expose partial success gaps.
Coding agents can pass tests by hardcoding examples, so green checks hide incorrect export-specific logic; targeted reviews catch it.
Playwright MCP lets coding agents review real browser behavior after frontend changes, reducing false completions.
AgentCore Session endpoints expose a coding agent's running Vite app for live browser previews via isolated sessions, auth, and allowed ports.
CQ2-CARE selects whole files for coding agents, avoiding compressed context that breaks patches. It beats compressors and retrievers on SWE-rebench.
Failure-conditioned learning trains coding agents from failed patches, improving Qwen3.5-4B to 52.8% Resolved@2 on SWE-bench Verified.
AI coding agents lower tool-operation skills but not output justification. Education should teach specifications, verification, and recovery.
Real coding agent sessions mix planning, execution, and review. SWE-TaskFlow aligns benchmarks to these workflows; multi-step interaction raises cost.
SWE-Duel has coding agents create and solve hidden bug tasks, making benchmarks self-scaling; GPT-5.5 solves best, Claude Opus 4.8 writes best.