# Cognition SWE-2 pricing: free in Devin Pro at $20 a month, no API price

> Devin's pricing page lists free SWE-2 use on the $20 Pro plan. Cognition has published no per-token price or API for the model.

URL: https://claudegpt.si/news/cognition-swe-2-pricing  
Published: 2026-11-05T02:53:46.988Z  
Author: ClaudeGPT Si Newsroom  
Section: AI Tools  
Topics: Cognition, Devin, SWE-2, Coding agents  
Publisher: ClaudeGPT Si

## Key takeaways

- As of November 5, 2026, Devin's Pro plan at $20 a month lists free use of SWE-2.
- Cognition's SWE-2 post and the Devin docs mention no API or per-token price for the model.
- Cognition says SWE-2 scores 73.0% on DeepSWE 1.1 and 92.8% on Terminal-Bench 2.1, and trails Fable 5.1 and GPT-6 Astra on Terminal-Bench 4.0.

SWE-2, Cognition's coding model, has no price of its own. As of November 5, 2026, you can't buy it per token: Cognition's [launch post](https://cognition.com/blog/swe-2) says it is available in Devin Desktop and the Devin CLI, and [Devin's pricing page](https://devin.ai/pricing) lists "Free use of SWE-2" on the $20-a-month Pro plan. The post and the pricing page don't mention an API for the model.

## What SWE-2 costs in Devin

These are the plans on [devin.ai/pricing](https://devin.ai/pricing), read on November 5, 2026.

| Plan | Price (as of November 5, 2026) | What the page lists |
| --- | --- | --- |
| Free | $0 | A light quota, limited model availability, unlimited inline edits and Tab completions |
| Pro | $20 a month | Higher quotas, frontier and open source models, Devin Cloud, free use of SWE-2, extra usage at API pricing |
| Max | $200 a month | Significantly higher quotas than Pro |
| Teams | $80 a month for the team, plus $40 a month per full dev seat | Up to 200 users, shared billing, admin dashboard, priority support |
| Enterprise | Custom | SSO, admin controls, dedicated deployment option |

The page doesn't give the size of the usage allowances, which it says refresh daily and weekly. It also doesn't say which tiers beyond Pro include free SWE-2, because its comparison table doesn't map that row clearly. Paid plans can buy usage beyond the allowance, billed at API pricing, but the page lists no SWE-2 rate.

Cognition's own cost claims are relative. The launch post says SWE-2 costs 64% less than Fable 5.1 on FrontierCode 1.1 Main, and that SWE-2 on medium effort costs 81% less than its predecessor SWE-1.7 on average on the same benchmark. It gives no dollar figure in the text.

## Is SWE-2 available through an API?

Not according to what Cognition has published. The launch post says SWE-2 is available in Devin Desktop and the Devin CLI, and rolling out to Devin Web and Fusion. [Devin's release notes](https://docs.devin.ai/release-notes/overview.md) say SWE-2 arrived on September 21, 2026 as a research preview in the agent selector, with Medium, High and Max reasoning effort. In Slack you pick it with `!swe2`. Enterprise admins turn it on in settings. The notes add that "behavior and availability may change."

The [Devin API reference](https://docs.devin.ai/api-reference/overview.md) describes REST endpoints for sessions and for managing an organization or enterprise. It doesn't mention SWE-2, model selection or pricing. So the only documented way to use SWE-2 is inside Devin.

## What SWE-2 is, and what Cognition claims

Cognition published SWE-2 on September 10, 2026. It says the model is post-trained from Kimi K3, a 2.8 trillion parameter model, using reinforcement learning, which trains the model on tasks where code either passes checks or doesn't. It comes in medium, high and max effort. Cognition reports mean steps per FrontierCode run of 53, 80 and 98 for the three levels.

The scores below are from Cognition's own table. The post doesn't report independent re-runs, and I found none in the sources I read.

| Benchmark | SWE-2 | Fable 5.1 | GPT-6 Astra | SWE-1.7 |
| --- | --- | --- | --- | --- |
| FrontierCode 1.1 Main | 50.0% | 50.9% | 53.3% | 42.0% |
| DeepSWE 1.1 | 73.0% | 67.4% | 74.1% | 37.7% |
| Terminal-Bench 2.1 | 92.8% | 91.4% | 89.9% | 81.5% |
| Terminal-Bench 4.0 | 27.3% | 55.8% | 57.9% | 7.6% |

The last row matters. On Terminal-Bench 4.0, SWE-2 scores less than half of what Fable 5.1 and GPT-6 Astra score. The cost claim holds where SWE-2 comes close, not on every hard task.

## SWE-2 Flash, a cheaper model announced November 4

On November 4, 2026, Cognition [announced SWE-2 Flash](https://cognition.com/blog/swe-2-flash), a smaller model built on Inkling-Small from Thinking Machines (276B total parameters, 12B active). Cognition says it costs one-fifth as much as SWE-2 and runs at about 200 tokens per second, roughly three times the 60 tokens per second it simulated for SWE-2. That speed figure comes from a replay of one Terminal-Bench 4.0 task, not a general measurement.

Cognition's table has Flash behind SWE-2 on DeepSWE 1.1 (60.1% against 73.0%) and Terminal-Bench 2.1 (89.2% against 92.8%). Flash is available in Devin Desktop and the Devin CLI, free through November 30. The post doesn't mention an API.

## What we don't know yet

- Any per-token price for SWE-2 or SWE-2 Flash.
- Whether SWE-2 will be offered outside Devin.
- How much usage the Pro plan's free SWE-2 covers, since the pricing page doesn't state allowances.
- How SWE-2 scores in tests Cognition didn't run.

For another look at AI coding editors and what they cost, see our [Windsurf vs Cursor comparison](https://claudegpt.si/news/windsurf-vs-cursor-comparison).

## Frequently asked questions

### Is SWE-2 available through an API?

Cognition's launch post says SWE-2 is in Devin Desktop and the Devin CLI, and rolling out to Devin Web and Fusion. It and the Devin API reference don't mention API access to the model.

### How much does Devin cost in 2026?

As of November 5, 2026, Devin's pricing page lists Free at $0, Pro at $20 a month, Max at $200 a month, Teams at $80 a month plus $40 per full dev seat, and custom Enterprise pricing.

### Is there a SWE-2 benchmark result from someone other than Cognition?

The scores in Cognition's launch post are its own. I found no independent re-runs in the primary sources read for this article.

### What is SWE-2 Flash?

Cognition announced SWE-2 Flash on November 4, 2026 as a smaller, faster coding model it says costs one-fifth as much as SWE-2. It is free in Devin Desktop and the Devin CLI through November 30.

## Sources

- [Introducing SWE-2: Pushing the Pareto Frontier](https://cognition.com/blog/swe-2), Cognition
- [Introducing SWE-2 Flash](https://cognition.com/blog/swe-2-flash), Cognition
- [Devin pricing](https://devin.ai/pricing), Cognition
- [Devin release notes](https://docs.devin.ai/release-notes/overview.md), Cognition
- [Devin API reference](https://docs.devin.ai/api-reference/overview.md), Cognition