Cognition released SWE-2 Flash, a coding model that it says runs at three times the speed of its SWE-2 model at one-fifth the price. The model shipped November 4 in Devin Desktop and Devin CLI, and Cognition is offering it free through November 30.
SWE-2 Flash is built on Thinking Machines' Inkling-Small, a mixture-of-experts model with 276 billion total parameters and 12 billion active. Thinking Machines released Inkling-Small's full weights on July 30, 2026. Cognition calls the result "a fully American coding model that sets a new bar for intelligence at low cost and high speed."
On Cognition's benchmark table, SWE-2 Flash lands close to its larger sibling. It scores 89.2 on Terminal-Bench 2.1 against 92.8 for SWE-2, and 45.5 on FrontierCode 1.1 Main against 50.0. On FrontierCode 2.0 it edges past SWE-2, 24.0 to 23.8.
Key facts
- 01Cognition released SWE-2 Flash on November 4, 2026, in Devin Desktop and Devin CLI.
- 02Cognition says the model runs at 3x the speed of SWE-2 at one-fifth the price.
- 03It is built on Thinking Machines' Inkling-Small, a 276B-parameter MoE model with 12B active parameters.
- 04SWE-2 Flash scores 89.2 on Terminal-Bench 2.1 versus 92.8 for SWE-2.
- 05The model is free in Devin through November 30; its weights are not public.
The bigger story in the table is how far Cognition moved the base model. On Terminal-Bench 4.0, Inkling-Small scores 0.9 and SWE-2 Flash scores 20.0. On DeepSWE 1.1 the jump is from 17.4 to 60.1, and on ExploitBench from 11.3 to 34.8, against 36.1 for SWE-2.
SWE-2 still leads on the hardest long-horizon tasks, scoring 27.3 on Terminal-Bench 4.0 and 73.0 on DeepSWE 1.1. Cognition says SWE-2 Flash outperforms all American open-weight models on leading evaluations, including models three times its size, a claim based on the company's own testing.
To show off the speed, Cognition's launch post replays a Terminal-Bench 4.0 run at a simulated 200 tokens per second for SWE-2 Flash versus 60 for SWE-2. The demo is a replay with transcripts trimmed to equal length, not a measured throughput benchmark.
Cognition is pitching SWE-2 Flash at high-volume work inside Devin: automations, code review, feature development and documentation updates.
Cognition has not published a per-token price, and no standalone API price list has been reported. Coverage of the launch differs on which Devin plans get the free window through November 30; Cognition's own post says only that the model is free through that date.
Cognition is keeping the weights closed for now. "We are not sharing the weights publicly yet," the company wrote, adding that it is building a program to securely share them with trusted industry partners and research labs, with more details promised in a few weeks.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
