Cognition just made its Devin coding agent cheaper and better at once, cutting prices up to 70% across modes while its Fusion harness climbed to 68.8 on the FrontierCode 1.1 benchmark, a combination that lands right after Cognition crossed $1 billion in annualized revenue.
Usually you get one or the other. A company drops its price because the product can't win on merit. Or it holds the line on price because the benchmark gains justify it. Cognition just did both to Devin at the same time, and it did so days after Bloomberg reported the AI coding startup topped $1 billion in annualized revenue, roughly double its run rate from four months earlier.
The price cuts are not a rounding error. Devin is now 30 to 40% cheaper in its Fusion and Normal modes, 15 to 20% cheaper in Ultra mode, and up to 70% cheaper for Devin Review, according to Cognition's own account of the changes. The company credits smarter model routing and caching, not a discount on quality. That claim gets a real number behind it: Devin Fusion shipped to Cognition's CLI on September 11. It scored 68.8 on the Extended portion of the FrontierCode 1.1 benchmark, at an average cost of $0.60 per task.
Fusion works by pairing a capable lead model with a cost-efficient sidekick, with Cognition drawing on its own SWE-2 model, Opus 5.5, the GPT-6 family and other models as needed. SWE-2 itself is post-trained from Moonshot AI's Kimi K3 at what Cognition describes as multi-trillion-parameter RL scale. It scored 50.0% on FrontierCode 1.1's main track, within a single point of Fable 5.1, at 64% lower cost. It also hit 92.8% on Terminal-Bench 2.1. The setup gives Cognition more room to cut prices without giving up the benchmark ground that justified Devin's premium in the first place.
This is the part that should worry Devin's rivals more than the benchmark number itself. Cognition just raised $2 billion at a $48 billion valuation this month, according to Bloomberg, and counts Nvidia, Citigroup and Mercedes-Benz as customers. A company that just doubled its revenue run rate in four months and closed a $2 billion round does not cut prices because it needs the cash. It cuts prices because it wants share, and it has just enough runway and enough of a benchmark lead to spend some of it buying that share now rather than later.
Cognition's SWE-2 Coding Agent Matches Rivals at a Quarter of the Price Cognition launched SWE-2 on September 10, 2026, claiming its new Devin coding agent scores within a point of Anthropic's Fable 5.1 on real-world benchmarks while costing up to 70% less to run. The company's own numbers show SWE-2 leads on cost but still trails frontier models on the hardest agentic tasks, landing the same week OpenAI faced... - cognition SWE-2 coding agent pricing and performance comparison - AI coding agent cost effectiveness versus anthropic models
GitHub Copilot and Cursor are the two names most exposed to that math. Copilot still leads on raw subscriber count: roughly 4.7 million paid users against Cursor's more than 1 million, based on an April 2026 IdeaPlan market survey. Copilot's $10-a-month Pro tier undercuts Cursor's $20 Pro plan by half. Cursor, now owned by SpaceX after a $60 billion all-stock deal that closed on August 14, has leaned into Composer 2 and cloud agent workflows rather than competing on price. Neither company sells an equivalent of Devin's per-task cost figure. That's exactly why Cognition publishing one, $0.60 a task on a named benchmark, changes the conversation. It gives enterprise buyers a number to hold every other vendor's agent against. That is the real pressure point. Seat-based pricing, the model both Copilot and Cursor still lean on, assumes a human logs in and does a fixed amount of work in a month. An autonomous coding agent doesn't work that way. It runs unattended, burns tokens on tasks of wildly different size, and its actual cost to the vendor swings with every model it calls underneath. Cognition's move is simple enough: cut price, publish a benchmark cost per task. It's a bet that usage-based, cost-per-task pricing is where this market ends up, and that whoever gets there first with the lowest verified number sets the reference point everyone else gets compared against.
None of this means Devin is now unambiguously the best coding agent on the market. FrontierCode 1.1 is one benchmark among several, and Cognition is the one publishing its own scores on it. But the direction is hard to miss: a startup that nearly doubled its valuation since May is now also the cheapest per-task option among the frontier coding agents, at least by its own measurement. For GitHub and Cursor's parent SpaceX, the question is no longer whether to compete with Devin on capability. It's whether they can afford to compete with it on price, too.
Also read: Shopify lets AI agents finish checkout on its two million stores. AMD buys Fei-Fei Li's World Labs for $8.2 billion in an all-stock deal. Shopify Opens Every Storefront to AI Agents With New WebMCP Tools
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.
Cognition buys Poke to plant its AI agent inside your text messages Cognition, the $26 billion startup behind autonomous coding agent Devin, has acquired Poke, the only AI agent Apple officially supports inside iMessage. The deal signals a strategic pivot toward ambient consumer AI embedded in SMS rather than standalone apps, as Poke logged over 100 million message exchanges in its first three months. - Cognition acquires Poke AI agent - AI agent Apple Messages integration