A new system called MemGuard, developed by researchers and posted on arXiv (2608.21867v1), improves LLM-agent memory reliability by treating verifier output as persistent lifecycle metadata, achieving…
Ornith-1.0, a June 2026 open-weight coding model, claims to learn by building its own harness during training, and its 9B model matches or beats Gemma4-31B on SWE-Bench Verified and Terminal-Bench 2.1…
LLM-as-a-Verifier, a framework for fine-grained agent feedback, reports that self-verification with DeepSeek V4 Flash outperforms Claude Fable 5 on Terminal-Bench 2.1, achieving 86.5% ± 1.1% Pass@1 fo…
Microsoft released MAI-Code-1.1-Flash on August 11th and put the coding model into production in GitHub Copilot, replacing the first version of its in-house coding model just 10 weeks after launch. Th…
Researchers propose SWE-Pruner Pro, a method that prunes tool outputs directly inside a coding agent by using the agent's own internal representations to label each line as keep or prune, saving up to…
Anthropic's Claude Fable 5, released June 9, 2026, leads the SWE-Bench Verified leaderboard with a 95% issue resolution rate, far ahead of Claude Opus 4.8 at 88.6%. The model's 1M-token context window…
NVIDIA released Nemotron 3 Super, a 120-billion-parameter hybrid model that scores 60.47% on SWE-Bench Verified, beating GPT-OSS-120B by nearly 20 points while delivering 2.2 times the inference throu…
Researchers introduced LLM-as-a-Verifier, a general-purpose verification framework that computes continuous scores from scoring token logits to determine solution correctness without additional traini…
Grok 4.5 and GPT-5.6 Sol both target agentic coding at competitive prices, with Grok 4.5 scoring in the high-50s on SWE-bench Verified and GPT-5.6 Sol scoring comparably, according to a comparison by …
Researchers introduced Single-rollout Asynchronous Optimization (SAO), a new reinforcement learning method for large language models that uses single-rollout sampling and double-side token-level clipp…
Anthropic released Claude Sonnet 5 on June 30, offering Opus-class agentic performance at 60% of the price, with introductory pricing of $2/$10 per million tokens expiring August 31. The model beats O…
Microsoft launched MAI-Code-1-Flash, a 137B-parameter sparse Mixture-of-Experts model with 5B active parameters, in GitHub Copilot on June 2, one day after GitHub switched to usage-based AI Credits bi…
DeepReinforce AI released Ornith 1.0, an open-source coding model family that scores 82.4 on SWE-Bench Verified, outperforming Claude Opus 4.7's 80.8, and runs the 35B variant locally on a single RTX …
Strands Labs released Simple Strands Agent (SSA), a minimal autonomous coding harness that achieves state-of-the-art results on SWE-Bench Verified, SWE-Bench Pro, and Terminal Bench 2 across 21 models…
Telnyx has made Moonshot AI's Kimi K2.6 model available for AI Assistants in the US region, offering developers on-network inference to reduce latency and simplify infrastructure. The model scores 58.…
StepFun released Step 3.7 Flash, a 198-billion parameter sparse MoE model, which achieves 76.3% on SWE-Bench Verified, matching 97% of Claude Opus 4.6’s coding performance at one-ninth the cost—$0.19 …