SWE-2
SWE-2CognitionProprietaryReasoningReleased Sep 10, 2026
Overall score
—
Not enough evidence to rank yet
Context window
1M
1,000,000 tokens
Price per 1M tokens
— in · — out
No price listed
Blended price
—
3:1 input:output tokens
Output speed
—
Not measured yet
Benchmark results (4)
As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.
Agentic & tool use
| Terminal-Bench 2.1 | 92.8 | #1 of 68 |
| Terminal-Bench 4.0 | 27.30 | #17 of 18 |
Coding
| DeepSWE | 73.0 | #12 of 50 |
| FrontierCode 1.1 Main | 50.0 | #5 of 16 |
Pricing by source
USD per 1M tokens.
| Source | Input | Output | Cached in |
|---|---|---|---|
| List price | — | — | — |
Cognition serves SWE-2 only inside Devin (Desktop and CLI at launch, rolling out to Devin Web and Fusion) and publishes no standalone API or per-token rate. The launch post cites per-task costs on FrontierCode 1.1 Main relative to other models, which are not a comparable token price.
Speed
Measured by us through OpenRouter (median of trials).
Not available on OpenRouter, so we cannot measure it yet.