{"rewrite":{"id":"r_8c4ae6c399c08f3e38cbdde1","clusterId":"c_21b2f3737d51e881e69e2346","slug":"cognition-releases-fusion-harness-claiming-up-to-39-percent-cost-cut","model":"deepseek-v4-flash","headline":"Cognition Releases Fusion Harness Claiming Up To 39 Percent Cost Cut","summary":"Cognition, the company behind the AI agent Devin, released Fusion, a harness that pairs a planning and review model with a cheaper execution model. Cognition advises GPT-6 Astra or Anthropic's the model as the planning model and the SWE-2 coding model as the executor, and states Fusion cuts costs by up to 39 percent versus Codex at comparable scores.","whyItMatters":"Cognition's pitch is that harness design, not model choice alone, decides the cost of an agent run, and it is now selling that harness inside Devin Desktop and CLI.","webCardHtml":"\u003cp\u003eFusion runs two agents in parallel with separate contexts and tools, exchanging only summaries, results, and feedback instead of the whole conversation. Cognition states that lets the pair make maximum use of prompt caching.\u003c/p\u003e\u003cp\u003eThe company advises a top model for planning and review and a cheaper one for execution, naming the SWE-2 coding model, announced September 11, 2026. Fusion ships in Devin Desktop and CLI.\u003c/p\u003e","blueskyPost":"Fusion's 39 percent saving comes from splitting one model into two roles, not from a cheaper model. Cognition's harness prices the planner at GPT-6 Astra and Anthropic's the model, then runs execution on SWE-2.","twitterPost":"Fusion's cost cut is an architecture choice: Cognition prices the planner at GPT-6 Astra and Anthropic's the model, and runs execution on the cheaper SWE-2.","threadsPost":"Fusion does not cut costs by shrinking the model. It splits the work: GPT-6 Astra or Anthropic's the model handles planning and review, and the cheaper SWE-2 coding model does the execution. Cognition is selling the split, not the discount.","newsletterBlurb":"Cognition released Fusion, a harness that pairs a top model for planning and review with a cheaper model for execution, with GPT-6 Astra or Anthropic's the model advised as the planner and SWE-2, announced September 11, as the executor. Cognition states Fusion holds cutting-edge-level scores while cutting costs by up to 39 percent versus Codex. Fusion ships in Devin Desktop and CLI.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260914-cognition-fusion/\",\"title\":\"Harness 'Fusion' That Draws Out the Performance of GPT-6 Astra and the model Arrives, Matching Performance When Used with Claude Code or Codex While Cutting Costs by Up to 39%\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4689,"outputTokens":575,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1789361981,"createdAt":"2026-09-14T04:56:19.000Z","publishedAt":"2026-09-14T04:58:26.000Z","updatedAt":"2026-09-14T04:58:26.000Z"},"cluster":{"id":"c_21b2f3737d51e881e69e2346","canonicalTitle":"GPT-6 AstraとClaude Fableの性能を引き出すハーネス「Fusion」が登場、Claude CodeやCodexで使う場合と同等の性能を示しつつコストを最大39％削減","representativeArticleId":"a_a2400481fb070b3cddc90bc6","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"GPT-6 Astra\",\"the model\",\"Fusion\",\"Devin\"],\"studios\":[\"Cognition\",\"Anthropic\"],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-09-14T03:56:00.000Z","lastSeenAt":"2026-09-14T03:56:00.000Z","updatedAt":"2026-09-14T04:58:21.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260914-cognition-fusion/","title":"GPT-6 AstraとClaude Fableの性能を引き出すハーネス「Fusion」が登場、Claude CodeやCodexで使う場合と同等の性能を示しつつコストを最大39％削減"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["GPT-6 Astra","the model","Fusion","Devin"],"studios":["Cognition","Anthropic"],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":null}