Playbooks · Coding tasks
Claude Fable 5.2 arrives in Devin Desktop and CLI
Written November 13, 2026
Fable 5.2 is now selectable in Devin Desktop and the Devin CLI. What Cognition's changelog says about the rollout.
What Cognition announced
Cognition's announcement, dated November 9, 2026, is short: Claude Fable 5.2 is now live in Devin. The post closes with the two doors into it — download Devin Desktop, or install the Devin CLI — which is the whole availability story the announcement tells. Fable 5.2 is a model a Devin user can select in the desktop editor and in the terminal tool, from the day Anthropic shipped the model.
That is worth stating plainly because model arrivals in agent tools often split: a model appears in a cloud product weeks before the local tools get it, or the other way round. Here Cognition names both of its surfaces in one announcement, on launch day. A developer who works in Devin Desktop and a developer who works from the CLI are looking at the same new option.
The one result Cognition leads with
The announcement's evidence is a single benchmark, and it is Cognition's own. On FrontierCode 2.0 — which Cognition describes as evaluating models on real-world engineering tasks, scored on quality and mergeability — it says Fable 5.2 takes the top spot at 49.1%. The comparisons it gives are Fable 5.1 at 32.4% and Opus 5.5 at 45.5%.
Cognition adds one cost observation: at low reasoning effort, Fable 5.2 already passes Fable 5.1's best result at less than a third of the cost per task. The benchmark's own notes, as Cognition publishes them, define the score as a weighted aggregate of rubric items, with solutions that miss blocking criteria scoring zero, and cost as the mean spend per rollout at list API pricing.
Read those numbers as what they are: the tool maker's own test, run its own way, published alongside the model's arrival in its products. They are a reason to try the model on your own work, not a ranking settled by a neutral party.
What the announcement does not say
- It names no plan gate. Nothing in the post limits Fable 5.2 to a tier, and nothing promises it to every tier either; the model picker in your own install is where that answer lives.
- It does not make Fable 5.2 the default. The claim is availability — the model is live and selectable — not that existing sessions change model on their own.
- It carries no Devin-specific pricing. The FrontierCode cost figures are benchmark accounting at list API prices, not a Devin price list.
Two surfaces, two working styles
The two named surfaces suit different habits. Devin Desktop is the editor, where model choice sits next to the code under review. The Devin CLI runs in a terminal on the developer's own machine and can hand work to Devin's cloud agent. Cognition's announcement covers both in one breath, so a team does not have to standardize on one surface to standardize on the model: the same Fable 5.2 option follows the developer from editor to terminal. That parity also makes a fair personal test easier — the same task, run once in each surface, shows whether the model or the surface is doing the work in any difference you see.
What to do
Update Devin Desktop or the Devin CLI, open the model picker, and run Fable 5.2 against a task you already know the shape of — a change you could review line by line. That comparison, on your repository, tells you more than the launch benchmark can. Developers who read about Fable 5.2 becoming the default Fable model in Claude Code should keep the two products separate: that default belongs to Claude Code's release, and Cognition's announcement makes no equivalent claim for Devin.
Related playbooks
- Agent-desktop drives apps via the access tree
- Antigravity CLI 1.5.3: subagent models, sandbox fix
- Antigravity CLI 1.5.4: ACP Auto mode
- Antigravity CLI 1.5.5: swarms in preview
Sources
Back to: Coding tasks playbooks hub · Also: Claude Code 2.1.321: Fable 5.2 is default · SWE-2 Flash: Devin's cheaper model, closed weights · Devin adds budgets, self-serve SSO and a usage API