Skip to content

Cognition SWE-1.7: Frontier Coding at $1.97 a Task

Cognition's SWE-1.7 Chases Frontier Coding on a Budget

Cognition shipped SWE-1.7 on July 8, 2026: a coding model landing near GPT-5.5 at about $1.97 per task. Here's what it claims, and where to be skeptical.

· · 3 min read
A modern computer chip mounted on an electronic circuit board

Quick Take

Cognition released SWE-1.7 on July 8, 2026, a coding model built inside the Devin harness that lands within a hair of GPT-5.5 on its own benchmarks while claiming to cost a fraction as much per task.

Another Week, Another Coding Model, and This One Is Interesting

Cognition, the company behind the Devin coding agent, shipped SWE-1.7 on July 8, 2026. In a year where a new frontier-ish model lands roughly every fortnight, most of them blur together. This one has a sharper pitch than usual: near-frontier coding quality at about $1.97 per task, which is a fraction of what the top proprietary models cost for comparable agentic work.

So what did they actually build, and should you care? Let me lay out the claims, then the caveats.

A blue printed circuit board photographed up close
Photo by Umberto on Unsplash

The Numbers Cognition Is Quoting

SWE-1.7 was RL post-trained on Kimi K2.7 Code, run inside the real Devin harness, and it's serving at around 1000 tokens per second on Cerebras. On Cognition's own FrontierCode 1.1 benchmark it scored 42.3, landing just 0.7 points behind GPT-5.5 and 4.2 behind Claude Opus 4.8. On SWE-Bench Multilingual it actually beat GPT-5.5, 77.8 to 76.8, though both sat behind Opus 4.8 at 84.4. On the independent Terminal-Bench 2.1 it managed 81.5, behind both frontier models but not by a mile.

The pattern is consistent: SWE-1.7 gets within a few points of the frontier on agentic coding while claiming a much lower cost per task. If those numbers hold in your workload, that's a real efficiency story, not a marketing one.

A close-up of a computer screen with many lines of code
Photo by Timothy Cuenat on Unsplash

Where I'd Be Careful

Here's the honest part. Every benchmark above is Cognition's own reporting. And when has a vendor ever published an eval that made its own model look bad? No independent lab has reproduced them yet, and vendor-run evals have a way of flattering the vendor. GPT-5.5 and Opus 4.8 still lead outright on the harder tests, so "matches the frontier" is a cost-adjusted claim, not a quality-parity one. I'd read "near-frontier at low cost" as the real message, and I'd verify it on my own tasks before trusting it in production.

The model is also tuned for long-horizon, asynchronous work inside Devin. That's a specific shape of problem, autonomous agent runs rather than quick inline completions, so it matters whether that's how you actually use a coding model day to day.

The Practical Take

If your AI spend is dominated by long agentic coding runs, SWE-1.7 is worth a pilot on the strength of the cost claim alone. Benchmark it against whatever you run now on your own repositories, not on FrontierCode, and watch cost per completed task rather than headline scores. For the wider picture of where models like this fit alongside the editors and agents you already use, our guide to practical AI coding tools is the place to start.

The short version: a genuinely interesting release with a sharp cost angle, and a set of benchmarks that deserve your own verification before you believe the headline.

Frequently Asked Questions

Is SWE-1.7 actually as good as GPT-5.5 or Claude Opus 4.8?
Close on some benchmarks, not on all, and the numbers are Cognition's own. On the vendor's FrontierCode 1.1 test SWE-1.7 scored 42.3, just 0.7 behind GPT-5.5 and 4.2 behind Claude Opus 4.8. On SWE-Bench Multilingual it edged out GPT-5.5 (77.8 to 76.8) but trailed Opus 4.8 at 84.4. On the independent Terminal-Bench 2.1 it hit 81.5, behind both. So it's genuinely near-frontier on cost-adjusted terms, but no independent lab has reproduced these results yet, so treat them as a starting point, not a verdict.
How much does SWE-1.7 cost to run?
Cognition quotes roughly $1.97 per task, which is the whole pitch: near-frontier coding at a fraction of what the top proprietary models charge for similar agentic workloads. It also runs fast, around 1000 tokens per second on Cerebras hardware. If your bill is dominated by long agentic coding runs, a model that lands close on quality at a meaningfully lower per-task cost is worth a serious pilot.
What is SWE-1.7 actually built for?
Long-horizon, asynchronous coding tasks inside Cognition's Devin harness, the same environment their customers use in production. The model was RL post-trained on Kimi K2.7 Code, and Cognition says it's more likely than the base model to investigate a bug's root cause and consider edge cases and adversarial inputs. In practice that means it's tuned for agent-style work that runs for a while unattended, not just quick single-shot completions.