@cognition
Makers of Devin, the first AI software engineer. We are an applied AI lab building end-to-end software agents. Join us: https://t.co/4Ss9hvpjRG
The FrontierCode leaderboard is now live: a dedicated page that tracks which models are writing code you’d actually merge. All scores — including Grok 4.5 and Inkling — are available, along with full methodology and sample tasks.
See the scores: cognition.com/frontiercode
Introducing Devin for Startups: $65k in credits to use Devin across Cloud, Desktop, and CLI. Apply today at the link below
In addition to credits, startups accepted will get access to exclusive events, merch, and white glove support. We’re hosting our first event in two weeks here: luma.com/devin-for-star…
Ship straight from the conversation. Devin in @Slack lets your team investigate issues, answer codebase questions, and start dev tasks without leaving the channel. Thanks for the feature, @Slack👇️ x.com/SlackHQ/status…
One year ago today, Cognition acquired Windsurf. In the year since, we've shipped dozens of new features, added hundreds of millions of dollars in ARR, and published frontier AI research. @ScottWu46 and @jeffwang share what we've accomplished in one year of building together.
As of today, Devin Fusion now incorporates Fable 5. Surprisingly, Fable 5 runs at a lower cost per task than Opus 4.8. Even though it’s a more expensive model, we observed cost efficiency improvements in areas like delegation and reasoning chains.
We found Fable works more like a competent manager delegating to an engineer, by writing comprehensive specs, giving detailed feedback, and trusting its sidekick with execution. Other models like Opus 4.8 tend to micromanage: they redo work, reread referenced files, and stuff their context window.
Yesterday we launched SWE-1.7 built on the open-source Kimi K2.7. Concerns about Chinese base models are real: K2.7 completed 87% of tasks that other models refuse over human-rights concerns. We trained SWE-1.7 specifically on trustworthiness, so it matches US models on evals.
When conversing in Simplified Chinese, leading open source models generally exhibit higher propaganda rates, alignment with CCP narratives, and refusals on politically sensitive questions.
GPT-5.6 is now available in Devin! On FontierCode 1.1 Extended, the GPT-5.6 family stands out for pairing strong scores with excellent cost efficiency. GPT 5.6 Sol reaches top performance at nearly half of the cost of the next best model.
Try it now in Devin Cloud, Devin Desktop, and Devin CLI. Read more at devin.ai/blog/gpt-5-6
Introducing SWE-1.7, the most capable model we’ve trained yet. It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s. RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale
SWE-1.7 was built on broad improvements in RL pipeline on top of a Kimi K2.7 base model. Our proprietary benchmark FrontierCode evaluates whether a model makes code you’d actually want to merge. SWE-1.7 advances the cost-performance Pareto curve on FrontierCode with a score of 42.3% and cost per task of $1.97 on our Main set.
We’ve made improvements to the FrontierCode methodology and are releasing FrontierCode 1.1 with clearer guidelines for fair internet use and refined grading criteria. Learn more about the changes on our blog: cognition.com/blog/frontier-…
Research, design, and math are three fields that tend to attract obsessives. Few have had a bigger impact on our imagination than legendary creator Grant Sanderson of @3blue1brown. We were lucky to host Grant at our office last week with Cognition's Head of Research @silasalberti and head of Design @josephhhhz for a fireside chat on the craft of making abstract ideas accessible, and how intelligence and design will evolve together. Events like this happen regularly at Cognition - if that sounds interesting you, you might be a great fit for our team. Check out our open careers here: t.co/ZsKXnw6E2f
Building on Devin's new capabilities, today we're announcing the Security Vulnerability Remediation Program: a tailored engagement to take your security backlog towards zero in six weeks. Remediation no longer has to compete with your product roadmap for engineering time.
Frontier models are collapsing the time from vulnerability discovery to exploitation. Findings that were once safe to deprioritize can no longer sit in the backlog. It's time to move from a reactive to proactive security posture.
Claude Fable 5 is available in Devin. You can use Fable 5 in Devin Cloud’s Ultra agent, our smartest and most capable agent, which excels at long-horizon tasks and debugging. Claude Fable 5 is also available in Devin Desktop and Devin CLI.
Try it in Devin! devin.ai
Introducing Devin Security Swarm A more cost effective and accurate way to find security vulnerabilities in complex codebases, based on a new architecture: Agentic MapReduce.
In testing, Devin Security Swarm found 36 of 50 real-world GHSA vulnerabilities at 30% lower cost per finding than the next most accurate alternative.
Claude Sonnet 5 is now available in Devin Desktop and Devin CLI. Sonnet 5 pairs frontier-level coding performance with a more affordable price point and outperforms Opus 4.8 on FrontierCode Extended.
On FrontierCode (Extended), our benchmark for real-world engineering tasks that grades mergeability and quality, Sonnet 5 scores 53.8% and has a 57.6% pass rate (higher than Opus 4.8). Note: These relative rankings may change slightly with coming adjustments to FrontierCode. t.co/yvKy9Ypeos
Watch Devin Fusion work: The sidekick model handles fetching information and skimming sources. The frontier model makes a plan for implementation, then delegates well-scoped sub-tasks to the sidekick with its own cached context. x.com/cognition/stat…
Conventional model routing sucks. It passes benchmarks but fails to write code you'd actually merge. Introducing Devin Fusion, a new hybrid-model harness for agentic coding. In testing, it reduces the cost of Fable-level intelligence by 35% and still feels good to use.
We developed FrontierCode to try to measure whether you’d actually merge a PR to production. Devin Fusion builds on this research while avoiding common pitfalls of model routing, like expensive cache misses and lack of generalization.
Most SOC2 audits take weeks of manual time. Read how @s16h_ and the @MetaviewAI team finished it in two days with Devin: x.com/s16h_/status/2…
Try Kimi K2.7 and GLM 5.2 for free in Devin Desktop and CLI x.com/devindesktop/s…
Devin tests its work before you review the PR. You review and approve the test plan. Then get back a screen recording with a visual checklist of step-by-step QA.
Read the docs for more details on testing: docs.devin.ai/work-with-devi…
Security review is now part of every Devin Review. Open a PR and Devin automatically finds the vulnerabilities scanners miss, explains each one, and drafts the fix.
Devin reasons across your whole codebase, so it catches business-logic flaws and chained findings that only surface when you understand how the app actually works. For example, Devin can catch a password-change endpoint that lets an unauthenticated request reset any user's password. The flaw only appears when you trace how the request moves through the auth layer, making it invisible to a pattern-matching tool.
We’re removing access to Claude Fable 5 in Cognition products following Anthropic’s latest announcement and the US government directive to suspend access. All other models remain available in Devin, including Claude Opus 4.8 and GPT-5.5. Devin Ultra mode will continue to use
Last week we introduced the AI Productivity Guarantee: a $10,000,000 bet that Devin delivers engineering value. Today, @ryanbai1412 walks you through how we calculate, benchmark, and guarantee it: x.com/ryanbai1412/st…
Congratulations to @axliu42 on receiving the ACM Doctoral Dissertation Award for his work on learning-theoretic foundations for understanding quantum systems. Allen joined Cognition's research team this week while continuing as a professor at NYU. We’re honored and excited to x.com/TheOfficialACM…
/handoff is one of the most powerful features of Devin CLI: close your laptop, your agents keep working from the cloud. Now, we've open sourced it. x.com/imjaredz/statu…
Two days left to apply for the Inference-Time Compute Hackathon, hosted by Cognition, @mercor_ai, @Etched, and @AnthropicAI, where you get: • 8x H100s per team • $100k+ in prizes • Dedicated Agents Track Build something that pushes the frontier.
@mercor_ai @Etched @AnthropicAI Apply here: luma.com/hncudfxb June 19-20 · SF
Claude Fable 5 is now available in Devin. Fable 5 earns the #1 spot on FrontierCode, our benchmark for real-world engineering tasks that grades mergeability and quality:
You can try Claude Fable 5 as part of Devin Cloud’s Ultra agent. Devin Ultra is our smartest and most capable agent, which excels at long-horizon tasks and debugging. We tuned the harness so Ultra costs only ~40% more than default Devin agent. Claude Fable 5 is also available
Introducing FrontierCode: a coding eval that raises the bar for difficulty & quality. Each task took 40+ hrs of work by leading open-source maintainers. Models write sloppy code that works but isn’t maintainable. Our eval is first to measure: would you actually merge this code?
20+ world-class open-source developers built realistic coding tasks on repos they maintain. They define what “mergeable” means in their repo. What does it take to measure mergeability? We use a mix of unit tests, rubrics and novel verifiers to assess correctness, test quality,
More than half of Cognition's first 50 employees were former founders. @ScottWu46 sat down with @breeves08 and @lux_capital to talk about correctness over comfort, staying focused on our mission, and what it takes to build an incredible team.
Thank you to @lux_capital for hosting. Full video: youtube.com/watch?v=3z1xNM…
AI should earn its keep. Introducing the AI Productivity Guarantee. If Devin delivers less engineering value than you’re paying for, Cognition will fund your usage until it does, up to $10 million. It’s time for the AI industry to stop maximizing tokens and start maximizing
We built a system to estimate whether an agent’s output was actually productive, and if so, how long a human engineer would have taken to do the same work. Validated on real engineers’ time estimates working in enterprise codebases. Read the blog from @ScottWu46:
At @harvey, the engineering team integrated Spectre — their internal background agent — into Devin Desktop. Now Spectre's organizational context can live on every engineer's laptop and flow across their favorite agents.
Bring your favorite agent into Devin Desktop using ACP: docs.devin.ai/desktop/acp
At @harvey, the engineering team integrated Spectre — their internal background agent — into Devin Desktop. Now Spectre's organizational context can live on every engineer's laptop and flows across their favorite agents.
Bring your favorite agent into Devin Desktop via ACP: docs.devin.ai/desktop/acp
Introducing Devin Desktop. Manage fleets of local and cloud agents from one surface. Plan, delegate, review, and ship without leaving your editor.
Devin Desktop is built on Windsurf, with all the features you know and love. x.com/windsurf/statu…
Join us, @mercor_ai, @Etched, and @AnthropicAI for a one-day hackathon in SF with a $50k top prize. Registrations close on 6/12. We can't wait what to see you build! x.com/mercor_ai/stat…
We've seen lots of enthusiasm for Devin's end-to-end testing capabilities in its virtual machine. In this post, @ido_pesok shares the technical details behind how we implement testing for verification at scale. x.com/ido_pesok/stat…
Claude Opus 4.8 is now available in Windsurf and Devin CLI x.com/windsurf/statu…
5/ Today, Devin is responsible for 89% of the PRs written at Cognition. Devin has dramatically accelerated our product roadmap, building and shipping new features like Devin Review, Auto-Triage, Managed and Scheduled Devins, Windsurf 2.0, and more.
6/ More details on our business and the rise of autonomous software engineering: cognition.ai/blog/series-d
