Synthszr Charts — die großen AI-Marken im Wettkampf ums Podium

SWE-1.6

Cognition · v1.6 · od 2026-04-07 · 7× · naposledy 30. 6. 2026

Momentum

SWE-1.6 is a proprietary AI model for software engineering agents developed by Cognition, released on April 7, 2026, as a generally available version within Windsurf (now rebranded as Devin Desktop). It was post-trained from scratch using reinforcement learning to jointly optimize raw benchmark performance and so-called "model UX" — efficient trajectories, parallel tool calls, and reduced looping. The model is exclusively available through Windsurf/Devin Desktop; local or standalone API use is not supported. It is offered in two speed tiers: free via Fireworks at 200 tok/s, and a paid Fast tier via Cerebras infrastructure at 950 tok/s.

Vývoj momenta

26.05.24.08.

Vlastnosti

Code Benchmark Score	SWE-Bench Pro: SWE-1.6 Preview improved on SWE-1.5 by >10% (11% per preview blog); SWE-1.6 Final achieves comparable performance to the preview model on SWE-Bench Pro, with 40% fewer turns
Deployment Model	Cloud-only (hosted service via Cognition/Windsurf); inference infrastructure: Free tier on Fireworks (GPU), Fast tier on Cerebras (wafer-scale chips); no self-hosting or local operation
Throughput Speed	Free tier: 200 tok/s (via Fireworks); Fast tier: up to 950 tok/s (via Cerebras) — per Cerebras approx. 5x faster than GPU-based free tier
Integration Focus	Exclusively integrated into Windsurf / Devin Desktop (Cascade or Devin local agent harness); proprietary model, cannot run in other IDEs or via a custom API
Parallel Tool Calls	Yes, explicitly supported and used significantly more often than in SWE-1.6 Preview; specifically encouraged via length penalty during training, reducing sequential tool calls and looping behavior
Price Tier	Free tier: free (for 3 months from release) at 200 tok/s via Fireworks; Fast tier: for paying users at 950 tok/s via Cerebras

Další produkty v této kategorii: Coding agenti

Zdroje (7)

Company Analysis: Cognition

As of 22/08/2026

SELLSynthszr Vote

On a data‑driven basis, the balance of evidence tilts toward SELL for new or marginal capital at current private‑market marks. Cognition’s fundamentals are undeniably strong—near‑$500M ARR, exceptional growth from 2024–2026, blue‑chip customers and a premier investor base—but the latest disclosed $26B valuation (and reports of talks at ~$40B) implies a revenue multiple in the ~50–80x range, far above even aggressive high‑growth SaaS benchmarks. Competitive intensity from Big Tech coding assistants, IDE‑native tools and multi‑agent platforms is rising, and Google‑Trends‑style indicators show that public interest has transitioned from explosive growth to a more stable, professional‑user‑driven baseline rather than continued acceleration. With limited transparency into long‑term retention, unit economics and true differentiation versus rivals, the current valuation leaves little room for execution missteps or market‑structure surprises. For investors with access to liquidity, taking profits or avoiding new exposure is prudent; only those with a venture‑style, long‑horizon risk appetite should consider selective participation, and even then with position sizes sized for potential downside volatility.

Key Takeaways

Cognition (maker of Devin) has scaled from a $10.2B valuation in September 2025 to a $26B post‑money valuation in its May 27, 2026 Series D, raising over $1B and more than doubling its value in eight months, with reports that it is already exploring a further round at ~$40B. (cognition.com)
The company disclosed ~$492M annualized run‑rate revenue as of May 2026, implying a revenue multiple of roughly 50–55x ARR at the $26B valuation—well above typical high‑growth SaaS but in line with the current premium for leading AI agents. (cognition.com)
Growth has been extremely rapid: ARR reportedly grew from about $1M in September 2024 to $73M by June 2025 and then to nearly $500M by May 2026, driven by >10x enterprise usage growth YTD and large customers such as Citi, Mercedes‑Benz, Goldman Sachs, Dell, Santander and major systems integrators like Infosys. (cognition.com)
Competitive and execution risks are material: Devin faces intense competition from Big Tech coding tools (e.g., GitHub Copilot, Claude Code, Google‑linked agents, Slack Code integrations) and specialized IDEs like Cursor, while some developers question real‑world reliability and security, including recent reports of product‑tier confusion and vulnerabilities after a June product split. (techcrunch.com)
Google‑Trends‑style and web‑traffic proxies suggest interest followed a hype spike around the March 2024 Devin launch and subsequent updates, then normalized into a stable but not rapidly rising baseline by mid‑2026; third‑party dashboards show Devin AI search interest modestly down (~low‑double‑digit) versus peak, while cognition.ai web traffic is steady in the low‑hundreds‑thousands of monthly visits, indicating STABLE rather than surging mainstream awareness. (trendspark-ai-plan.lovable.app)

Action Ideas

SELL

At a private valuation of ~$26B (with talk of ~$40B) on disclosed ~$492M ARR, Cognition trades at ~50–80x revenue depending on which forward run‑rate investors underwrite, embedding expectations of sustained hypergrowth, durable technical leadership and eventual dominance of the AI‑coding‑agent category. While recent ARR growth and blue‑chip customer logos are impressive, the competitive field is crowded with well‑funded incumbents (Microsoft/GitHub Copilot, Anthropic’s Claude Code, Google‑linked agents, Cursor) and emerging multi‑agent platforms, and switching costs for developers remain relatively low. Any slowdown in enterprise expansion, pricing pressure from bundled Big Tech offerings, or negative publicity around reliability/security (e.g., concerns about over‑hyped demos, product‑tier confusion, or vulnerabilities after the June product split) could compress multiples sharply from today’s venture‑style levels. For investors able to sell in secondary markets or who participated at much lower entry valuations, the risk‑reward now skews toward taking profits and recycling capital into cheaper AI exposure.

Horizon: 24 mo.

BUY

For high‑risk, long‑duration investors who can access primary or secondary shares, Cognition offers leveraged exposure to what may become a foundational layer in software development workflows if autonomous coding agents reach mass adoption. The company has demonstrated exceptional revenue growth—from low‑single‑digit millions to nearly $500M ARR in under two years—alongside rapid enterprise uptake and partnerships with major corporates and integrators, suggesting Devin is moving beyond hype into real production use. If Devin (and successors) become a standard co‑engineer embedded across IDEs, CI/CD pipelines and enterprise platforms, the addressable market could justify multi‑billion‑dollar revenue potential, making today’s valuation less extreme on a 5–7‑year view. The technical talent density, strong investor syndicate, and ongoing product iteration (e.g., Devin 2.x, desktop, government and enterprise offerings) further support a thesis that Cognition can remain among the top few winners in the agentic‑coding space. For investors comfortable with binary‑like outcomes and illiquidity, small, venture‑style allocations may be attractive despite headline multiples.

Horizon: 60 mo.

HOLD

For existing shareholders who entered at earlier rounds (e.g., sub‑$10B valuations), maintaining exposure while avoiding new capital commitments appears reasonable. The company’s fundamentals—near‑$500M ARR, marquee customers, and strong investor backing—support the current decacorn status, but the jump to $26B (and rumored $40B) leaves little room for disappointment. Google‑Trends‑style indicators and web‑traffic data show that public interest has cooled from 2024’s viral peak into a more stable, professional‑user‑driven baseline, suggesting the easy narrative tailwinds are behind it. With the competitive landscape still in flux and limited transparency into unit economics, churn and net expansion, the risk‑reward for adding at current marks is not compelling; however, fully exiting could forgo upside if Devin consolidates a durable lead as enterprises standardize on a small set of agents. A wait‑and‑see stance—monitoring renewal behavior, competitive benchmarks and any future funding at higher or lower valuations—balances these considerations.

Horizon: 18 mo.

Google Trends · → stable

Across the last two years, Google‑Trends‑style and related indicators for Cognition/Devin show a classic hype‑cycle pattern: a sharp global spike in search interest around the March 2024 launch of Devin as the “first fully autonomous AI software engineer,” followed by elevated but choppier interest through subsequent product updates and funding rounds, and then a normalization into a lower but persistent baseline by mid‑2026. Third‑party dashboards tracking Devin AI as a keyword report that its interest score is now modestly below peak (low‑double‑digit percentage decline) even as broader AI‑agent and coding‑assistant queries continue to rise, while web‑analytics data for cognition.ai show steady traffic in the low‑hundreds‑thousands of monthly visits with a healthy share of branded search. Taken together, these data points suggest that general consumer/developer curiosity has cooled from the initial viral moment, but sustained professional and enterprise engagement has kept overall interest roughly STABLE rather than clearly rising or collapsing.

Contrarian Insights

• Despite the dominant media narrative that Cognition’s success is driven by broad developer adoption and viral hype, third‑party traffic and sentiment data suggest a more concentrated, enterprise‑heavy story: cognition.ai’s monthly visits are in the low‑hundreds‑thousands range and largely direct or branded search, while community sentiment trackers show only moderate, recently improving buzz. This implies that a relatively small number of high‑value enterprise deployments may be driving a disproportionate share of ARR, making growth more sensitive to a few large accounts than headline valuation figures imply. (es.semrush.com)
• Many observers frame Devin as either a looming replacement for software engineers or an over‑hyped demo destined to fade, but emerging usage patterns and integrations point to a more nuanced equilibrium: Devin increasingly appears as one specialized agent among several (e.g., alongside Claude Code, Copilot, Cursor and Slack Code), embedded in multi‑agent workflows rather than acting as a monolithic “AI engineer.” If this multi‑tool reality persists, Cognition’s upside may come less from monopolizing coding and more from owning high‑value niches (complex multi‑step tasks, regulated‑industry workflows, government/enterprise integrations), which could still support strong economics but may not justify the most aggressive valuation scenarios currently discussed in private markets. (digitalproductsdp.com)