Cognition announced SWE-2 Flash on November 4, 2026, making a new coding model available in Devin Desktop and CLI, its command-line tool. The company advertises free use through November 30, giving developers and small software teams a window to assess it for everyday engineering work.
The offer concerns model use, not a free Devin subscription. The announcement does not specify which account types qualify or detail service limits, so it should not be read as a promise of unlimited access for everyone.
What the free offer covers
Devin’s standard pricing page lists a limited Free tier, Pro at US$20 a month and Max at US$200 a month. Those are subscription prices, not Flash rates. The page does not confirm whether Free-tier accounts qualify for the new offer.
The quota guide says free models do not consume the normal usage allowance. That exemption is different from unlimited service and does not establish free access to other models or cloud work.
The model documentation still names SWE-2—not SWE-2 Flash—in its promotional and post-promotion pricing notes. Those terms should not be transferred to Flash. A Flash-specific price after November 30 is not established in these materials; the documentation points users to the in-app selector for current pricing and availability.
Before relying on the offer, look for the exact SWE-2 Flash entry in your account’s model selector. In Devin CLI, /model opens that selector. For enterprise users, administrators can restrict the available models, so a public launch does not override a team’s access policy.
Faster generation is not a finished job
Cognition describes Flash as one-fifth the price and three times the speed of SWE-2. These are company comparisons, not independently established productivity savings. Its speed illustration uses simulated playback of shortened, equal-length transcripts; it does not demonstrate how long equivalent software tasks take from start to finish.
The practical implication is to compare time to a reviewed, usable change, including retries and human checking, rather than how quickly text appears. A small code change or documentation update fits the routine work Cognition describes. The promotional window offers a chance to make that comparison; the speed claim alone is not a reason to replace an existing tool.
No public self-hosting release
Flash is built on Thinking Machines’ Inkling-Small, whose weights are publicly available. Weights are the trained model files used to run it. That access does not carry over to Cognition’s derivative: SWE-2 Flash’s weights are not public.
Cognition says it is developing controlled weight sharing for partners and researchers, not offering a general download now. Teams that need to host Flash themselves should not plan around access that has yet to be announced. For background on the separately released base model, see RohitAI’s Inkling-Small explainer.