Stories
Page 2 of 6 · newest source date first.
Latest stories
- Cloud
Devin Cloud Agents add Mac VM support for iOS, macOS apps
Cognition announced that Devin Cloud Agents now support Mac virtual machines.
- Desktop
Cognition launches SWE‑2 model: 50% FrontierCode score, 64% lower cost claim
Cognition's SWE‑2 model, built on a 2.8 T‑parameter Kimi‑K3 base, hits 50% on FrontierCode and is offered via Devin Desktop and CLI only.
- Cognition
Independent test finds Devin Fusion scores 62 on Coding Agent Index, matching Claude Fable 5.1
Benchmark shows Devin Fusion reaches a 62 CI score, equal to Claude Fable 5.1, while offering lower cost and faster execution.
- CLI
Devin CLI adds Fusion model‑pairing for lower‑cost frontier performance
The new Fusion harness lets developers combine Fable and Astra models in the Devin CLI, delivering up to 39% efficiency gains on coding benchmarks.
- Cognition
Cognition: Introducing SWE-2: Pushing the Pareto Frontier
Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort levels, pushing the Pareto f…
- Cognition
Cognition Expands with Dioxus Acquisition
Cognition acquires Dioxus, bolstering its portfolio with new technologies
- Cognition
Devin Voice launches with voice commands
Cognition announced Devin Voice, a conversational interface that lets developers speak commands and have full software tasks executed.
- Cognition
Cognition Achieves Breakthrough in RSA-260 Factoring
Cognition's new GPU lattice siever reduces factoring costs by 90% compared to previous state of the art
- Cognition
Cognition Achieves Breakthrough in RSA-260 Factoring
Cognition's GPU lattice siever reduces factoring costs by 90% compared to previous methods
- Cognition
Cognition Raises $2B in Series E
Cognition announces Series E funding at $48B valuation to build the future of software engineering
- Desktop
Devin Desktop 3.9.1018 drops Cascade, migrates users to Devin Local
The 3.9.1018 update eliminates Cascade, makes Devin Local the sole agent, and introduces UI tweaks and bug fixes.
- Desktop
GPT-6 Astra debuts on Devin with lower cost than Fable 5
Cognition has made the GPT-6 Astra model available on Devin Cloud, Desktop, and CLI.
- Cloud
Fable 5.1 in Devin: the cache-read cut that flipped the frontier economics
Cognition shipped Claude Fable 5.1 across Devin Cloud, Desktop, and CLI on August 31 with a 54% per-task cost reduction versus Fable 5 — $2.68 versus $5.84 on FrontierCode. The win comes from Anthropic's 75% cache-read discount, making Fable-class intelligence economical for the workloads Cognition had kept on Opus. Fusion routing upgraded to use Fable 5.1 as orchestrator, matching Fable scores at 47% lower cost.
- CLI
Devin CLI adds /model command for model selection and cost comparison
Cognition’s official X post says the Devin CLI now offers a /model command for selecting models, tuning reasoning effort, and comparing costs.
- Cognition
The partnership we missed: Fiserv is deploying Devin against core banking modernization
Fiserv and Cognition announced a strategic partnership on May 28 to put Devin to work modernizing core banking platforms. It resurfaced in high-reach syndication this month — and it slots neatly into the partnerships machine Cognition has been assembling since.
- Cognition
SpaceX reportedly tried to buy Cognition. Scott Wu says it is not for sale
Bloomberg reported SpaceX approached Cognition days after closing its $60 billion Cursor acquisition. CEO Scott Wu called the story inaccurate and said the company isn't for sale — while Bloomberg separately reports early funding talks at a $40 billion valuation.
- Desktop
GPT‑5.6 Sol gets a further 20% price cut in Devin Desktop and CLI
Cognition said OpenAI cut Sol's prices by another 20% for three months, for a 76% discount off list price through October 3.
- Desktop
Coming back to Devin Desktop? Adaptive mode is the setting returners keep choosing
A lapsed Windsurf user came back, picked Adaptive to stretch his grandfathered credits, and shipped two open-source PRs — with one complaint. Adaptive is the Desktop-side sibling of the routing bet Cognition made with Fusion, and it's been quietly maturing since April.
- Desktop
Cognition adds Grok 4.6 to Devin Desktop and CLI with XHigh reasoning mode
The update makes the third‑party model available on both the desktop application and command‑line interface, and Cognition says it outperforms GPT‑5.6 Sol.
- Cognition
Cognition is quietly building a partnerships machine
The LTM deal put Devin inside a managed security service aimed at 80% of an enterprise's CVE backlog — the first of five joint offerings. Add a reported senior partnerships hire from Brex and the FedRAMP push, and a distribution strategy comes into focus.
- CLI
Third-party toolmakers now treat Devin CLI as a platform target
Skill frameworks are shipping Devin CLI support alongside Claude Code and Codex, an open-source plugin hands tasks to Devin from competing agents, and Desktop 3.7.16 shipped the plumbing — synced plugins, a Subagents section, layered permissions. The harness is becoming a platform.
- CLI
Plan on the Desktop, ship from the terminal, finish in the cloud
Power users have converged on a three-surface Devin workflow — and their loudest feature request is the one seam that's still manual. Here's the handoff pattern, what the docs actually support today, and where the gap is.
- Cloud
Fusion's first field reports: the sidekick math is starting to check out
Cognition claims Devin Fusion holds frontier scores at 35–60% lower cost. Six weeks into the preview, Japanese practitioner posts and a high-reach routing panel are treating the sidekick pattern as the reference design — here's what's published, what's user-reported, and what's still open.
- Cloud
Japan's engineering teams are writing the best Devin playbook. Here's what it says
Zenn's production trial, a team that spent six months learning to spend its leftover ACUs, Devin Review pointed at the Dependabot queue, and a designer shipping refactors from Slack — the JP practitioner cluster keeps publishing the operational detail everyone else skips.