
?28pc×ßÊÆ×î×¼Õ¹Íû?ΪÄãÌṩ28pc×ßÊÆ×î×¼Õ¹ÍûAPP°²×¿°æÏÂÔØ£¬£¬£¬£¬£¬£¬£¬ÀúÊ·°æ±¾¡¢¾É°æÏÂÔØ£¬£¬£¬£¬£¬£¬£¬Éó²é×îÐÂ28pc×ßÊÆ×î×¼Õ¹ÍûÊÖ»ú°æÏÈÈÝ¡¢Ó¦ÓýØÍ¼¡¢ÍøÓÑ̸ÂÛ£¬£¬£¬£¬£¬£¬£¬Àû±ã¿ì½ÝµÄ½«°²×¿°æ28pc×ßÊÆ×î×¼Õ¹ÍûÓ¦ÓÃÃâ·ÑÏÂÔØµ½ÊÖ»ú¡£¡£¡£¡£¡£¡£¡£¡£
ÊÇÒ»¿îºÜÊǴ̼¤µÄðÏÕ½âÃÕÓÎÏ·£¬£¬£¬£¬£¬£¬£¬ÔÚÕâ¸öÓÎÏ·ÖУ¬£¬£¬£¬£¬£¬£¬Äã¿ÉÒÔ×ÔÓɵØÔÚ·èÈËÔºÖÐ̽Ë÷£¬£¬£¬£¬£¬£¬£¬Óжà¸ö·¿¼äÐèÒª½âËø£¬£¬£¬£¬£¬£¬£¬Ã¿¸ö·¿¼ä¶¼ÓÐ×Ô¼ºµÄ¹ÊʺÍÌØ±ðµÄÃÕÌ⣬£¬£¬£¬£¬£¬£¬Ö𲽽⿪Õâ±³ºóµÄÕæÏ࣬£¬£¬£¬£¬£¬£¬Ï²»¶µÄÅóÙ½Ó´ýǰÀ´ÏÂÔØÌåÑé¡£¡£¡£¡£¡£¡£¡£¡£
ÊÇÒ»¿îÕ½ÂÔËþ·ÀÓÎÏ·¡£¡£¡£¡£¡£¡£¡£¡£Íæ¼Ò·ÖΪÁ½ÅÉ£¬£¬£¬£¬£¬£¬£¬ÊÎÑÝÄãµÄ½ÇÉ«Íê³ÉʹÃü£¬£¬£¬£¬£¬£¬£¬ ѰÕÒÍâÐÇÁÔÎ£¬£¬£¬£¬£¬£¬×¢ÖØÎ±×°³ÉÈËÀàµÄ³ðÈË£¬£¬£¬£¬£¬£¬£¬ÖÆ×÷Ëþ·ÀÉèÊ©£¬£¬£¬£¬£¬£¬£¬ÁÔɱÍâÐÇÈË£¬£¬£¬£¬£¬£¬£¬ÊÎÑÝÄãµÄ½ÇÉ«£¬£¬£¬£¬£¬£¬£¬¸ü¿ìµÄ¾ÙÐÐÉú³¤¡£¡£¡£¡£¡£¡£¡£¡£Ò»³¡À´×ÔÉñÃØÐÇÇòµÄÉúËÀÖ®Õ½£¬£¬£¬£¬£¬£¬£¬¼¤¸ÐÈËÐĵÄÓÎÏ·½Ú×࣬£¬£¬£¬£¬£¬£¬½«Äã´øÈëÒ»¸ö³äÂúÉñÃØËþ·ÀÕ½ÌìÏ¡£¡£¡£¡£¡£¡£¡£¡£
ÊÇÒ»¿îÊ®·ÖÓÐȤµÄÐж¯Õ½¶·ÓÎÏ·£¬£¬£¬£¬£¬£¬£¬ÒÔÆäÁ¢ÒìÐÔµÄÍæ·¨ºÍ¸»ºñ¶à²ÊµÄÌì϶øÍÑÓ±¶ø³ö¡£¡£¡£¡£¡£¡£¡£¡£ÔÚÕâ¸öÁîÈË×ÅÃÔµÄÓÎÏ·ÖУ¬£¬£¬£¬£¬£¬£¬Íæ¼ÒÊÎÑÝÂíͰÈË£¬£¬£¬£¬£¬£¬£¬¼ç¸º×ÅÊØÎÀÌìÏÂÃâÊÜа¶ñµÄ³ðÈËÈëÇÖµÄÖ÷ҪʹÃü¡£¡£¡£¡£¡£¡£¡£¡£
ÊÇÒ»¿î³äÂú´Ì¼¤µÄÉúÑÄðÏÕÓÎÏ·£¬£¬£¬£¬£¬£¬£¬¿ÖÁúÖÖÀ฻ºñ£¬£¬£¬£¬£¬£¬£¬ÕæÊµ»¹ÔÁË¿ÖÁúʱ´úµÄÉú̬ÇéÐΣ¬£¬£¬£¬£¬£¬£¬Íæ¼Ò½«ÊÎÑÝÒ»¸ö±»À§ÔÚÒ°ÍâµÄÈË£¬£¬£¬£¬£¬£¬£¬±ØÐèÔÚÕâ¸öΣÏյĵط½ÉúÑÄÏÂÀ´£¬£¬£¬£¬£¬£¬£¬¸ÐÐËȤµÄÍæ¼Ò¿ÉÒÔÏÂÔØÊÔÊÔ¿´Å¶¡£¡£¡£¡£¡£¡£¡£¡£
ÊÇÒ»¿î»·çÇåÎúϸÄåµÄÏñËØÃ°ÏÕÓÎÏ·£¬£¬£¬£¬£¬£¬£¬Íæ¼ÒÐèÒªÍê³ÉһϵÁÐʹÃüÒÔ½âËøÐÂÇøÓòºÍÌôÕ½£¬£¬£¬£¬£¬£¬£¬ÕâЩÐ×Ã͵Ľ©Ê¬ÃǻᳯÄãÃÇÌᳫ½ø¹¥£¬£¬£¬£¬£¬£¬£¬Ð¡Í¬°éÃÇÏëÒªÉúÑÄÏÂÈ¥µÄ»°£¬£¬£¬£¬£¬£¬£¬¾ÍÐèÒª½«ËüÃÇÈ«¶¼»÷°Ü²ÅÐÐŶ£¬£¬£¬£¬£¬£¬£¬Ò²¿ÉÒÔ×ÔÖÆÎäÆ÷£¬£¬£¬£¬£¬£¬£¬¿ìÀ´ÏÂÔØÕ½¶·°É¡£¡£¡£¡£¡£¡£¡£¡£
ÊÇÒ»¿î¿ì½Ú×àµÄ´òÂÝË¿ÓÎÏ·£¬£¬£¬£¬£¬£¬£¬ÔÚÏÞÖÆÊ±¼äÄÚ¾¡¿ÉÄÜ¶àµØÍê³É´òÂÝË¿¶¤µÄʹÃü£¬£¬£¬£¬£¬£¬£¬Ëæ×Źؿ¨µÄÌáÉý£¬£¬£¬£¬£¬£¬£¬ÄѶÈÒ²»áÒ»Ö±ÌáÉý£¬£¬£¬£¬£¬£¬£¬ÔÚÓöµ½Ò»Ð©½ÏÁ¿ÄÑÒÔÍê³ÉµÄ¹Ø¿¨Ê±£¬£¬£¬£¬£¬£¬£¬Íæ¼ÒÐèÒªÔËÓÃһЩ¼¼ÇÉÀ´Íê³ÉʹÃü¡£¡£¡£¡£¡£¡£¡£¡£
×ªÔØ×Ô£º½×Ô¾Ðdz½
Ïò¸÷ÈËÏÈÈÝÎÒÃÇ×îж˵½¶ËÉî¶ÈÑо¿ÖÇÄÜÌåÄ£×Ó£ºstep-deepresearch¡£¡£¡£¡£¡£¡£¡£¡£
![]()
ËüÄÜ×öʲô£¿£¿£¿£¿£¿£¿£¿
»á˼Ë÷£º¶®ÍýÏë¡¢»á·´Ë¼¡¢»¹ÄÜ×ÔÖ÷ÑéÖ¤ÐÅÏ¢µÄÕæÎ±£¬£¬£¬£¬£¬£¬£¬²»µ«ÊÇÊý¾ÝÅÀ³æ¡£¡£¡£¡£¡£¡£¡£¡£
¸üרҵ£ºÍŽá¸ßÖÊÁ¿ËÑË÷ api£¬£¬£¬£¬£¬£¬£¬ÄÚÖà 2000 Íò+ ¸ßÖÊÁ¿Îĵµ¿âºÍ 600+ ȨÍþÕ¾µãË÷Òý£¬£¬£¬£¬£¬£¬£¬¹ýÂ˵ÍÐÅÔ´ÐÅÏ¢×ÌÈÅ¡£¡£¡£¡£¡£¡£¡£¡£
¸üÖǻۣºÆæÒìµÄÔ×ÓÄÜÁ¦ÑµÁ··¨£¬£¬£¬£¬£¬£¬£¬ÈÃÄ£×Ó°Ñ¡°×¨¼ÒÍ·ÄÔ¡±ÄÚ»¯ÔÚ¹Ç×ÓÀï¡£¡£¡£¡£¡£¡£¡£¡£
ÔÚÕë¶ÔÕæÊµÖØ´ó³¡¾°µÄ adr-bench ÆÀ²âÖУ¬£¬£¬£¬£¬£¬£¬step-deepresearch ÌåÏÖ³ö¼«Ç¿¾ºÕùÁ¦£¬£¬£¬£¬£¬£¬£¬ÔÚ¶à¸öά¶ÈÉÏÖÜÈ«ÓâÔ½º£ÄÚÍâÒ»Ïß deepresearch ²úÆ·ºÍÄ£×Ó£¬£¬£¬£¬£¬£¬£¬°üÀ¨»ùÓÚǧÒÚ¼¶ÉõÖÁÍòÒÚ¼¶²ÎÊý moe ¼Ü¹¹µÄ gemini deepresearch¡¢openai deepresearch µÈÄ£×Ó¡£¡£¡£¡£¡£¡£¡£¡£
¶ø step-deepresearch ½ö»ùÓÚµ¥ÖÇÄÜÌå¼Ü¹¹ÊµÏÖ¡£¡£¡£¡£¡£¡£¡£¡£
![]()
ÔÚPTÊÓѶ(ÖйúÇø)¹ÙÍøÊÖÒÕ±¨¸æÖÜÈ«¹ûÕæºó£¬£¬£¬£¬£¬£¬£¬Á¬Ã¦Òý·¢º£ÄÚÍâÉçÇøÈÈÁÒÌÖÂÛ£¬£¬£¬£¬£¬£¬£¬²¢ÈëÑ¡ huggingface daily papers¡£¡£¡£¡£¡£¡£¡£¡£
![]()
![]()
ÏÖÔÚ£¬£¬£¬£¬£¬£¬£¬ÎÒÃÇ¿ªÆô api Äڲ⣬£¬£¬£¬£¬£¬£¬½Ó´ý¸÷ÈËÓ»Ô¾ÌåÑ飡
ÂÛÎĵص㣺https://arxiv.org/pdf/2512.20491
github£ºhttps://github.com/stepfun-ai/stepdeepresearch
api Äڲ⣺https://wvixbzgc0u7.feishu.cn/share/base/form/shrcn8cp78pjgkjvvih2c3ef3cc
¹Ù·½Ö÷Ò³£ºhttps://www.stepfun.com/deep-research-invitation
step-deepresearch ÔÚ¶àÏîȨÍþ»ù×¼²âÊÔÖж¼µÖ´ïÈ«Çò¶¥¼âˮƽ¡£¡£¡£¡£¡£¡£¡£¡£
ÔÚ research rubrics ÉÏ£¬£¬£¬£¬£¬£¬£¬step-deepresearch µÃ·Ö 61.42%£¬£¬£¬£¬£¬£¬£¬½ö´ÎÓÚ gemini deepresearch£¨Ô¼ 63.69·Ö£©£¬£¬£¬£¬£¬£¬£¬ÇÒÓâÔ½ÁË openai deepresearch¡£¡£¡£¡£¡£¡£¡£¡£
![]()
ÔÚ¡°ÒýÓÃÖÊÁ¿¡±ºÍ¡°ÏàͬÖÊÁ¿¡±Á½¸öά¶ÈÉϵִïÁËÐÐÒµ×î¸ßˮƽ¡£¡£¡£¡£¡£¡£¡£¡£
![]()
ÔÚÕë¶ÔÕæÊµÑо¿³¡¾°µÄ adr-bench ÆÀ²âÖУ¬£¬£¬£¬£¬£¬£¬step-deepresearch µÄÈËÀà elo ÆÀ·ÖÔÚ¶à¸öά¶È´¦ÓÚµÚÒ»Ìݶӡ£¡£¡£¡£¡£¡£¡£¡£
ÌØÊâÊÇÔÚÓë gemini deepresearchµÈ¶¥¼âÄ£×ÓµÄÖ±½Ó²©ÞÄÖУ¬£¬£¬£¬£¬£¬£¬Æä¡°Ê¤+ƽ¡±Âʸߴï67.1%£¬£¬£¬£¬£¬£¬£¬Ö¤ÊµÎúÆäÌìÉúµÄ±¨¸æÖÊÁ¿ÒѵִïÄ¿½ñ×îÏȽøË®Æ½¡£¡£¡£¡£¡£¡£¡£¡£
![]()
step-deepresearch ÒýÓÃ×îÐÂÐÐÒµ¶¯Ì¬£¨step-gui¡¢¶¹°üÊÖ»úµÈ£©£¬£¬£¬£¬£¬£¬£¬À´×ÔÉϰÙÌõÌõ¸ßÖÊÁ¿ÐÅÔ´£¬£¬£¬£¬£¬£¬£¬ÄÚÈÝÁýÕֵײãÊÖÒÕ¡¢Ó¦Óá¢ÌôÕ½ÓëÕùÒ飬£¬£¬£¬£¬£¬£¬¿Í¹ÛÖÜÈ«ÆÊÎöÒéÌ⣬£¬£¬£¬£¬£¬£¬µÖ´ï±ê×¼µ÷Éî¶ÈÑб¨¸æË®Æ½¡£¡£¡£¡£¡£¡£¡£¡£
![]()
¡øÉÏÏ»¬¶¯Éó²éÏêϸÄÚÈÝ
step-deepresearch µÄÖ÷ÒªÁ¢Òì¿É¹éÄÉΪ3·½Ã棺
»ùÓÚÔ×ÓÄÜÁ¦µÄÊý¾ÝºÏ³ÉÕ½ÂÔ
½¥½øÊ½ÖÇÄÜÌåѵÁ··¶Ê½
×Ô½¨ adr-bench ÆÀ¹Àϵͳ
ÎÒÃǽ«Éî¶ÈÑо¿ÄÜÁ¦²ð·ÖΪËĸöÔ×ÓÄÜÁ¦£¬£¬£¬£¬£¬£¬£¬²¢Õë¶ÔÐԵغϳÉÊý¾Ý¾ÙÐÐÇ¿»¯¡£¡£¡£¡£¡£¡£¡£¡£
ÍýÏëÓëʹÃü²ð½â£ºÄܽ«Ä£ºý¡¢ºë´óµÄÓû§ÐèÇó²ð½âΪ¿ÉÖ´ÐеÄ×ÓʹÃü£¬£¬£¬£¬£¬£¬£¬²¢Æ¾Ö¤ÇéÐη´Ï춯̬µ÷½â·¾¶¡£¡£¡£¡£¡£¡£¡£¡£
Éî¶ÈÐÅÏ¢ËÑË÷£º¾ß±¸¶àÌøÍÆÀíÄÜÁ¦£¬£¬£¬£¬£¬£¬£¬ÄÜÔÚÐÅÏ¢²»ÍêÕûʱ¾ÙÐС°×Ô¶¯ÍØÆËÐÐ×ß¡±£¬£¬£¬£¬£¬£¬£¬ÍÚ¾òÒþ²ØÊµÌå¡£¡£¡£¡£¡£¡£¡£¡£
·´Ë¼ÓëÑéÖ¤£ºÓµÓС°×Ô¾À´í¡±ºÍ¡°ÊÂʵºË²é¡±ÄÜÁ¦£¬£¬£¬£¬£¬£¬£¬ÄÜʶ±ð×ÔÉí¹ýʧ²¢Çø·ÖÍøÂçÐÅÏ¢µÄÕæÎ±£¬£¬£¬£¬£¬£¬£¬Í¨¹ý¿çÔ´Ñé֤ȷ±£Âß¼ÑÏÃÜ¡£¡£¡£¡£¡£¡£¡£¡£
±¨¸æÌìÉú£ºÍ¨¹ýÖÐÆÚѵÁ·£¨mid-training£©Ñ§Ï°×¨¼Òд×÷Æø¸Å£¬£¬£¬£¬£¬£¬£¬²¢Ê¹Óà sft È·±£±¨¸æÑÏ¿á×ñÕÕÍýÏë½á¹¹ºÍÒýÓù淶¡£¡£¡£¡£¡£¡£¡£¡£
PTÊÓѶ(ÖйúÇø)¹ÙÍøÖÇÄÜÌåѵÁ··¶Ê½Ö÷Òª°üÀ¨Èý¸ö½×¶Î£º
agentic mid-training£ºÔÚԤѵÁ·ºÍ΢µ÷Ö®¼ä¼ÓÈë mid-training£¬£¬£¬£¬£¬£¬£¬Í¨¹ý 32k ºÍ 128k Á½¸ö½×¶ÎµÄÉÏÏÂÎĵ÷Àí£¬£¬£¬£¬£¬£¬£¬×¢ÈëÔ×ÓÄÜÁ¦£¬£¬£¬£¬£¬£¬£¬Ê¹Ä£×ÓÄÚ»¯¡°ÏÂÒ»²½Ðж¯¡±µÄ¾öÒéÂß¼£¬£¬£¬£¬£¬£¬£¬¶ø·Ç¼òÆÓµÄ¡°Õ¹ÍûÏÂÒ»¸ö token¡±¡£¡£¡£¡£¡£¡£¡£¡£
sft£º×ÅÖØÓÚ³¤³Ì¾öÒé¹ì¼£µÄºÏ³ÉÓëÁìÓòÊÊÅ䣬£¬£¬£¬£¬£¬£¬Ç¿»¯ÒâͼÃ÷È·¡¢ÍýÏëÖ´Ðм°ÑÏ¿áÒýÓÃÃûÌõÄ×ñÕÕ£¬£¬£¬£¬£¬£¬£¬½â¾öÄ£×ÓÔÚ³¤Ê±¼äÑо¿Ê¹ÃüÖÐÈÝÒס°·ÖÉñ¡±»ò¡°ÃÔ·¡±µÄÎÊÌâ¡£¡£¡£¡£¡£¡£¡£¡£
rl£ºÒýÈë checklist-style judger ½±ÀøÉè¼Æ£¬£¬£¬£¬£¬£¬£¬½«ÖØ´óµÄ±¨¸æÖÊÁ¿ÆÀ¹Àת»¯ÎªÏ¸Á£¶ÈµÄÐźţ¬£¬£¬£¬£¬£¬£¬½øÒ»²½ÓÅ»¯³¤³Ì¾öÒéµÄ³°ôÐÔ¡£¡£¡£¡£¡£¡£¡£¡£
ÔÚϵͳ¼Ü¹¹ÉÏ£¬£¬£¬£¬£¬£¬£¬ÎÒÃǽÓÄɵ¥ÖÇÄÜÌå react ¼Ü¹¹£¬£¬£¬£¬£¬£¬£¬×èÖ¹ÖØ´ó¶àÖÇÄÜϵһÇÐÐ×÷´øÀ´µÄϵͳÈßÓà¡£¡£¡£¡£¡£¡£¡£¡£
![]()
±ðµÄ£¬£¬£¬£¬£¬£¬£¬ÎªÁËÌî²¹ÖÐÎÄÉî¶ÈÑо¿ÆÀ¹ÀµÄ¿Õȱ£¬£¬£¬£¬£¬£¬£¬ÎÒÃÇ»¹¹¹½¨ÁË adr-bench£¬£¬£¬£¬£¬£¬£¬º¸ÇÉÌÒµµ÷ÑС¢Õþ²ßÆÊÎö¡¢Èí¼þ¹¤³ÌµÈ 9 ´óÁìÓò£¬£¬£¬£¬£¬£¬£¬°üÀ¨Í¨ÓúÍרҵ£¨Ö´·¨¡¢½ðÈÚ£©Á½¸öά¶È¡£¡£¡£¡£¡£¡£¡£¡£¸Ã»ù×¼²âÊÔ²»µ«¹Ø×¢ËÑË÷Ч¹ûÊÇ·ñ׼ȷ£¬£¬£¬£¬£¬£¬£¬Ô½·¢¹Ø×¢Ñо¿Àú³ÌÖеÄÂß¼ÑÏÃÜÐÔ¡¢Òâͼʶ±ðÉî¶ÈÒÔ¼°³¤³Ì¾öÒéµÄ³°ôÐÔ¡£¡£¡£¡£¡£¡£¡£¡£
½Ó´ýת·¢£¬£¬£¬£¬£¬£¬£¬µ«Çë×¢Ã÷À´ÓÉ¡°ÉϺ£¾ÐÅί¡±
ÉϹۺÅ×÷ÕߣºÉϺ£¾ÐÅί
28pc×ßÊÆ×î×¼Õ¹Íû>