[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fT_MC7EfxxtxLd3kpSYLptKUlZec8VDaHpKA3Zu4Cx4I":3},{"article":4,"related":19},{"id":5,"slug":6,"title":7,"seo_title":7,"description":8,"keywords":9,"content":10,"category":11,"image_url":12,"source_guid":13,"published_at":14,"created_at":15,"updated_at":16,"source_url":17,"source_name":18},1330,"google-ships-gemini-38-flash-its-third-budget-model-in-6-weeks","Gemini 3.8 Flash raised capability and cost per task","Google's September Gemini 3.8 Flash release improved benchmark scores but raised measured task costs, making successful completion the key upgrade test.","[\"Gemini 3.8 Flash\",\"Google\",\"AI coding\",\"inference costs\",\"model evaluation\"]","\u003Cp>Google released Gemini 3.8 Flash with stronger reported coding performance and introductory token prices matching its predecessor, according to \u003Ca href=\"https:\u002F\u002Fthe-decoder.com\u002Fgemini-3-8-flash-is-googles-third-budget-model-in-six-weeks-while-frontier-models-remain-mia\u002F\" rel=\"noopener noreferrer\">the-decoder.com’s September 2, 2026 report\u003C\u002Fa>. The release included a general-purpose model and a restricted cybersecurity version.\u003C\u002Fp>\n\u003Cp>Google reported a DeepSWE v1.1 software engineering score of 73.7 percent, up from 65.3 percent for Gemini 3.7 Flash. Separately, independent benchmarking platform Artificial Analysis scored the new model at 59 on its Intelligence Index, versus 56 for its predecessor. Those results support a benchmark improvement, not a guarantee that a particular coding workflow will improve.\u003C\u002Fp>\n\u003Cp>The cost evidence complicates the upgrade. Artificial Analysis measured $0.58 per task for 3.8 Flash, compared with $0.40 for 3.7 Flash. Google attributed some capability gains to additional reasoning and repeated tool calls. Launch pricing was $0.75 per million input tokens and $3.75 per million output tokens, with regular prices of $1.50 and $7.50 scheduled for January 2027.\u003C\u002Fp>\n\u003Cp>For a team choosing between the models, the useful acceptance test is cost per satisfactory result. A comparison should count whether a coding task passes the team’s checks, how many retries it requires and how much reviewer intervention it needs. The reported cost per benchmark task does not establish that outcome. Higher spending could be justified by fewer failed attempts, but the source supplies no workload-specific evidence that this happens.\u003C\u002Fp>\n\u003Cp>The scheduled price increase also makes the evaluation date consequential. Teams considering an integration should price the same measured workload under both the introductory and regular rates before deciding. Google’s continued support for 3.7 Flash leaves room for a selective migration: require evidence that 3.8’s additional reasoning earns its cost on the tasks being moved.\u003C\u002Fp>\n\u003Cp>\u003Cem>Correction, September 19, 2026: This article was revised to attribute reported claims, remove unsupported statements and clarify the limits of the available evidence.\u003C\u002Fem>\u003C\u002Fp>","AI & Machine Learning","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1788394374372-39oiiq5e356.png","f7e114e24f3c4a5f929257cd81291b5741627be857aface8224c7482e78ffd5d","2026-09-02T16:59:29.000Z","2026-09-03T00:12:54.811Z","2026-09-19T19:28:34.746Z","https:\u002F\u002Fthe-decoder.com\u002Fgemini-3-8-flash-is-googles-third-budget-model-in-six-weeks-while-frontier-models-remain-mia\u002F","the-decoder.com",[20,27,34,41],{"id":21,"slug":22,"title":23,"description":24,"category":11,"image_url":25,"published_at":26},1366,"nvidia-releases-flora-for-robot-video-labels-and-rewards","Nvidia releases Flora for robot video labels and rewards","Nvidia's Flora combines robot video annotation and reward scoring. Its reported error detection gains come with false alarms and projected compute costs.","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1792973586727-g8o8rhrfkba.webp","2026-10-26T00:13:06.924Z",{"id":28,"slug":29,"title":30,"description":31,"category":11,"image_url":32,"published_at":33},1365,"coding-harness-study-finds-higher-bills-uncertain-gains","Coding harness study finds higher bills, uncertain gains","A study finds coding harness costs can triple without clear accuracy gains. Its results give teams a way to weigh prompt overhead against task completion.","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1792889312113-dpcnm8ft60v.webp","2026-10-25T00:48:32.358Z",{"id":35,"slug":36,"title":37,"description":38,"category":11,"image_url":39,"published_at":40},1360,"minimax-m31-flash-cuts-task-costs-as-benchmark-scores-rise","MiniMax M3.1 Flash cuts task costs as benchmark scores rise","MiniMax's M3.1 Flash improves benchmark scores at lower reported task costs. For AI teams, the decision turns on accuracy, uncertainty handling and actual sp...","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1792477333643-5zoq22ue5f7.webp","2026-10-20T06:22:13.877Z",{"id":42,"slug":43,"title":44,"description":45,"category":11,"image_url":46,"published_at":47},1358,"deepseeks-flash-update-makes-a-case-for-cheaper-coding-agents","Deepseek’s Flash update makes a case for cheaper coding agents","Deepseek’s Flash update claims stronger coding results at lower task costs. Its internal benchmarks give developers a reason to test, not a migration verdict.","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1792282370679-befpubjmnl.webp","2026-10-18T00:12:51.052Z"]