https://devotd.net/en/news/harness2-recursive-agent-harness/devotd.en2026-10-20T23:56:26ZHarness²: Automatic harness improvement, higher benchmark scoreshttps://devotd.net/ja/news/harness2-recursive-agent-harness/devotd.ja2026-10-20T23:56:26ZHarness²：ハーネスの自動改善、ベンチマーク性能の向上https://devotd.net/ko/news/harness2-recursive-agent-harness/devotd.ko2026-10-20T23:56:26ZHarness²: 하네스 자동 개선, 벤치마크 성능 상승https://devotd.net/zh-hans/news/harness2-recursive-agent-harness/devotd.zh-cn2026-10-20T23:56:26ZHarness²：自动改进智能体运行框架，提升基准性能https://devotd.net/en/news/deepseek-deepinfer-open-source-inference-framework/devotd.en2026-10-20T11:22:52ZDeepInfer: Offline inference, service componentshttps://devotd.net/ja/news/deepseek-deepinfer-open-source-inference-framework/devotd.ja2026-10-20T11:22:52ZDeepInfer：オフライン推論、サービス構成要素https://devotd.net/ko/news/deepseek-deepinfer-open-source-inference-framework/devotd.ko2026-10-20T11:22:52ZDeepInfer: 오프라인 추론, 서비스 구성요소https://devotd.net/zh-hans/news/deepseek-deepinfer-open-source-inference-framework/devotd.zh-cn2026-10-20T11:22:52ZDeepInfer：离线推理，服务组件https://devotd.net/en/news/ollama-0-40-7-rc0-responses-speculative-pplx/devotd.en2026-10-20T07:26:12ZOllama v0.40.7-rc0: Reasoning and draft handling improvements, decider model supporthttps://devotd.net/ja/news/ollama-0-40-7-rc0-responses-speculative-pplx/devotd.ja2026-10-20T07:26:12ZOllama v0.40.7-rc0：推論・ドラフト処理の改善、判定モデルのサポートhttps://devotd.net/ko/news/ollama-0-40-7-rc0-responses-speculative-pplx/devotd.ko2026-10-20T07:26:12ZOllama v0.40.7-rc0: 추론·초안 처리 개선, 판정 모델 지원https://devotd.net/zh-hans/news/ollama-0-40-7-rc0-responses-speculative-pplx/devotd.zh-cn2026-10-20T07:26:12ZOllama v0.40.7-rc0：改进推理与草稿处理，支持判定模型https://devotd.net/en/news/minimax-m31-flash/devotd.en2026-10-20T04:24:35ZMiniMax M3.1 Flash: agent benchmark results, claimed 70% task cost reductionhttps://devotd.net/ja/news/minimax-m31-flash/devotd.ja2026-10-20T04:24:35ZMiniMax M3.1 Flash：エージェント評価の成績、タスクコスト70%削減の主張https://devotd.net/ko/news/minimax-m31-flash/devotd.ko2026-10-20T04:24:35ZMiniMax M3.1 Flash: 에이전트 평가 성적, 작업 비용 70% 절감 주장https://devotd.net/zh-hans/news/minimax-m31-flash/devotd.zh-cn2026-10-20T04:24:35ZMiniMax M3.1 Flash：智能体评测成绩、任务成本降低70%的主张https://devotd.net/en/news/llama-cpp-iq-k-quantization-proposal/devotd.en2026-10-20T03:20:38Zllama.cpp PR #30715: IQ*_K Quantization, CPU Inferencehttps://devotd.net/ja/news/llama-cpp-iq-k-quantization-proposal/devotd.ja2026-10-20T03:20:38Zllama.cpp PR #30715：IQ*_K量子化、CPU推論https://devotd.net/ko/news/llama-cpp-iq-k-quantization-proposal/devotd.ko2026-10-20T03:20:38Zllama.cpp PR #30715: IQ*_K 양자화, CPU 추론https://devotd.net/zh-hans/news/llama-cpp-iq-k-quantization-proposal/devotd.zh-cn2026-10-20T03:20:38Zllama.cpp PR #30715：IQ*_K 量化，CPU 推理https://devotd.net/en/news/physical-plausibility-evaluator-screening/devotd.en2026-10-20T00:00:38ZNVIDIA Physical Plausibility Evaluator: Single-video screening, customizable decision criteriahttps://devotd.net/ja/news/physical-plausibility-evaluator-screening/devotd.ja2026-10-20T00:00:38ZNVIDIA Physical Plausibility Evaluator：単一動画の選別、独自の判定基準https://devotd.net/ko/news/physical-plausibility-evaluator-screening/devotd.ko2026-10-20T00:00:38ZNVIDIA Physical Plausibility Evaluator: 단일 영상 선별, 맞춤형 판정 기준https://devotd.net/zh-hans/news/physical-plausibility-evaluator-screening/devotd.zh-cn2026-10-20T00:00:38ZNVIDIA Physical Plausibility Evaluator：单视频筛选，定制判定标准https://devotd.net/en/news/apertus-1-5-early-fusion-multimodal-long-context/devotd.en2026-10-19T21:43:55ZApertus 1.5: expanded multimodal input, generational performance gains and evaluation disclosurehttps://devotd.net/ja/news/apertus-1-5-early-fusion-multimodal-long-context/devotd.ja2026-10-19T21:43:55ZApertus 1.5：マルチモーダル入力の拡張、前世代からの性能向上と評価の公開https://devotd.net/ko/news/apertus-1-5-early-fusion-multimodal-long-context/devotd.ko2026-10-19T21:43:55ZApertus 1.5: 멀티모달 입력 확장, 전세대 성능 개선과 평가 공개https://devotd.net/zh-hans/news/apertus-1-5-early-fusion-multimodal-long-context/devotd.zh-cn2026-10-19T21:43:55ZApertus 1.5：多模态输入扩展、较上一代的性能提升与评估公开https://devotd.net/en/news/gemini-4-argon-github-copilot/devotd.en2026-10-19T21:18:51ZGemini 4 Argon: General availability, per-token pricinghttps://devotd.net/ja/news/gemini-4-argon-github-copilot/devotd.ja2026-10-19T21:18:51ZGemini 4 Argon：正式提供、トークン別料金https://devotd.net/ko/news/gemini-4-argon-github-copilot/devotd.ko2026-10-19T21:18:51ZGemini 4 Argon: 정식 제공, 토큰별 요금https://devotd.net/zh-hans/news/gemini-4-argon-github-copilot/devotd.zh-cn2026-10-19T21:18:51ZGemini 4 Argon：正式推出，按 token 计费https://devotd.net/en/news/llama-cpp-b11724-cuda-flash-attn-warp-sync-racecheck/devotd.en2026-10-19T09:21:33Zllama.cpp b11724: CUDA race fix, warp synchronizationhttps://devotd.net/ja/news/llama-cpp-b11724-cuda-flash-attn-warp-sync-racecheck/devotd.ja2026-10-19T09:21:33Zllama.cpp b11724：CUDAの競合修正、warp同期https://devotd.net/ko/news/llama-cpp-b11724-cuda-flash-attn-warp-sync-racecheck/devotd.ko2026-10-19T09:21:33Zllama.cpp b11724: CUDA 경쟁 수정, warp 동기화https://devotd.net/zh-hans/news/llama-cpp-b11724-cuda-flash-attn-warp-sync-racecheck/devotd.zh-cn2026-10-19T09:21:33Zllama.cpp b11724：修复 CUDA 竞争，warp 同步https://devotd.net/en/news/h3-character-swap-lora-v1-5/devotd.en2026-10-19T05:25:20ZH3 Character Swap LoRA V1.5: Improved appearance retention, validation scope and editing limitshttps://devotd.net/ja/news/h3-character-swap-lora-v1-5/devotd.ja2026-10-19T05:25:20ZH3 Character Swap LoRA V1.5：人物の外見維持を改善、検証範囲と編集の限界https://devotd.net/ko/news/h3-character-swap-lora-v1-5/devotd.ko2026-10-19T05:25:20ZH3 Character Swap LoRA V1.5: 인물 외형 유지 개선, 검증 범위와 편집 한계https://devotd.net/zh-hans/news/h3-character-swap-lora-v1-5/devotd.zh-cn2026-10-19T05:25:20ZH3 Character Swap LoRA V1.5：人物外观保持改进，验证范围与编辑局限https://devotd.net/en/news/sglang-0-5-22-apple-silicon-instanttensor-pd-disaggregation/devotd.en2026-10-19T03:42:36ZSGLang v0.5.22: More Mac execution options, improved model loading and disaggregated servinghttps://devotd.net/ja/news/sglang-0-5-22-apple-silicon-instanttensor-pd-disaggregation/devotd.ja2026-10-19T03:42:36ZSGLang v0.5.22：Macの実行経路を拡大、モデル読み込みと分離サービングを改善https://devotd.net/ko/news/sglang-0-5-22-apple-silicon-instanttensor-pd-disaggregation/devotd.ko2026-10-19T03:42:36ZSGLang v0.5.22: 맥 실행 경로 확대, 모델 로딩·분리 서빙 개선https://devotd.net/zh-hans/news/sglang-0-5-22-apple-silicon-instanttensor-pd-disaggregation/devotd.zh-cn2026-10-19T03:42:36ZSGLang v0.5.22：扩展 Mac 运行路径，改进模型加载与分离式服务https://devotd.net/en/news/nemo-gym-siemens-fuse-eda-agent-verifier-turns-tokens/devotd.en2026-10-19T01:29:44ZNVIDIA NeMo Gym: Verification-driven instruction improvements, fewer chip design agent callshttps://devotd.net/ja/news/nemo-gym-siemens-fuse-eda-agent-verifier-turns-tokens/devotd.ja2026-10-19T01:29:44ZNVIDIA NeMo Gym：検証に基づく作業指示の改善、チップ設計エージェントの呼び出し削減https://devotd.net/ko/news/nemo-gym-siemens-fuse-eda-agent-verifier-turns-tokens/devotd.ko2026-10-19T01:29:44ZNVIDIA NeMo Gym: 검증 기반 작업 지침 개선, 칩 설계 에이전트 호출 절감https://devotd.net/zh-hans/news/nemo-gym-siemens-fuse-eda-agent-verifier-turns-tokens/devotd.zh-cn2026-10-19T01:29:44ZNVIDIA NeMo Gym：基于验证改进任务指引，减少芯片设计智能体调用https://devotd.net/en/news/ollama-cloud-fly-io-sprites-connector-gateway-key-isolation/devotd.en2026-10-19T01:21:06ZOllama Cloud: Sprites connectivity support, gateway API key injectionhttps://devotd.net/ja/news/ollama-cloud-fly-io-sprites-connector-gateway-key-isolation/devotd.ja2026-10-19T01:21:06ZOllama Cloud：Sprites接続に対応、ゲートウェイでAPIキーを付与https://devotd.net/ko/news/ollama-cloud-fly-io-sprites-connector-gateway-key-isolation/devotd.ko2026-10-19T01:21:06ZOllama Cloud: Sprites 연결 지원, 게이트웨이 API 키 주입https://devotd.net/zh-hans/news/ollama-cloud-fly-io-sprites-connector-gateway-key-isolation/devotd.zh-cn2026-10-19T01:21:06ZOllama Cloud：支持 Sprites 连接，网关注入 API 密钥