On July 9, 2026, Cognition AI said it had made OpenAI's latest model family, GPT-5.6, available inside its autonomous software engineer Devin, citing a mix of strong performance and cost efficiency on its own benchmark.
July 9, 2026 · Cognition AI · Devin
GPT-5.6 Lands Inside Devin — Top Coding Scores at Half the Cost
Cognition AI has rolled OpenAI's new three-model GPT-5.6 family into its autonomous software engineer, citing top performance and cost efficiency on its own production-grade coding benchmark, FrontierCode 1.1 Extended.
3
models in the family:
~½
the cost of the next best model, at top performance (Sol)
150
real-repo tasks in FrontierCode 1.1 Extended
Same performance, half the cost
Terra matches GPT-5.5-level results — for roughly half the price. Cost drawn to scale.
Sol
Most capable flagship
Top performance at ~half the next best model. Same ACU consumption rate as GPT-5.5.
Terra
Balanced perf & cost
GPT-5.5-level performance at about half the price.
Luna
Fastest & most affordable
Lowest cost per task in the family.
What FrontierCode measures
Not just "does it pass tests" — but "is it mergeable as a real production PR."
Real repos
Quality rubric
Fair use
Works well
Operates independently like a junior engineer
Saves time on PR creation & multi-step implementations
Useful code review support
Real limits
Can get stuck in loops or make wrong assumptions
Needs frequent guidance on complex tasks
Review still requires human checks
The open question: large-scale usage data after the GPT-5.6 integration is still limited — how far the mergeability-focused benchmark wins carry into real-world development remains to be seen.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…