Claude Code quietly swapped its default model on September 1, 2026. If you have not touched the /model picker in a few weeks, you are probably already running on Claude Fable 5.1 instead of Sonnet 5, and the CLI is defaulting to High effort while it does it. Here is what actually changed, what did not, and whether the switch is worth paying attention to.
Key Takeaways
- Claude Code made Claude Fable 5.1 the default model on September 1, 2026, running at High effort, while Claude Cowork and claude.ai default to Medium effort for the same model.
- Fable 5.1 keeps Fable 5's headline API rates but cuts cache read cost by 75%, from $1.00 to $0.25 per million tokens, lowering costs up to 45% on cache heavy agentic sessions.
- Benchmarks jumped sharply: Terminal-Bench-Science 0.1 more than doubled to 52.6%, and Terminal-Bench 4.0 reached 55.8%, ahead of Claude Opus 5's 52.3%.
- Teams that pin a model by name or run Claude Code through an unconfigured gateway stay on the older Fable 5, since the generic `fable` and `best` aliases need an explicit update to resolve to 5.1.
- For product managers, the upgrade quietly strengthens root cause analysis and multi step follow through on cache heavy work like discovery synthesis and PRD reviews, a good moment to revisit Claude Code for PMs.
Learn this hands-on
Go from idea to a live product with real users. Build and ship your first SaaS with Claude Code in a live cohort, with auth, payments and deployment done by the end. Join the Ship Your First SaaS with Claude Code cohort.
What Actually Happened on September 1
Anthropic released Claude Fable 5.1 and its restricted-access counterpart, Claude Mythos 5.1, on September 1, 2026. The same day, Claude Code 2.1.257 shipped and made claude-fable-5-1 the default Fable model across the CLI, replacing Claude Fable 5. In Claude Code specifically, the model now defaults to High effort. Claude Cowork and claude.ai, by contrast, default to Medium effort, so the same model behaves differently depending on where you call it from.
Fable 5.1 and Mythos 5.1 are the same underlying model shipped with different levels of safeguards. Fable 5.1 is generally available through Claude.ai, the Claude API, AWS, Google Cloud, and Microsoft Azure. Mythos 5.1 stays limited to vetted organizations in cybersecurity and life sciences.
There is one rollout wrinkle worth knowing about: gateway deployments that have not been configured for Fable 5.1 would reject requests naming it directly, so Claude Code 2.1.257 keeps the generic fable and best aliases resolving to Fable 5 for Claude apps gateway sessions until an operator updates their configuration. If you want Fable 5.1 specifically, select it by name in /model.
What Changed From the Previous Default
Sonnet 5 had been Claude Code's default since it launched with its native 1M-token context window (see our Claude Sonnet 5 breakdown for what that context window unlocked). The move to Fable 5.1 is not a context-window story, it is a cost and capability story on Anthropic's Mythos-class model line.
The clearest change is pricing. Fable 5.1 keeps the same headline API rates as the original Fable 5, $10 per million input tokens and $50 per million output tokens, but cuts the cost of a cache read from $1.00 to $0.25 per million tokens, a 75% reduction. Anthropic estimates that alone lowers costs by roughly 25% for typical workloads and by as much as 45% for highly agentic, cache-heavy tasks, which describes a large share of what Claude Code actually does across a long coding session: rereading the same files, the same context, the same tool definitions, turn after turn.
The benchmark gains are just as notable. On Terminal-Bench-Science 0.1, Fable 5.1 scored 52.6% against Fable 5's 24.7%, more than double. On Terminal-Bench 4.0 it reached 55.8%, ahead of both Fable 5 (42.0%) and, in third-party testing, Claude Opus 5 (52.3%). It also improved on Humanity's Last Exam (60.9% vs. 57.8%) and CursorBench 3.2.0 (73.4% vs. 70.5%). Anthropic points to particular strength in identifying the root cause of software issues and carrying complex, multi-step coding tasks through to completion, which is exactly the kind of long-horizon agentic work Claude Code sessions tend to involve.
If you remember when Fable 5 was a free preview back in July, this is the same model family maturing into something cheaper and sharper, not a new experiment.
See It in Action
Rather watch than read? I put the new default to the test here:
Is the Switch Worth It?
For most Claude Code users, this is not a decision at all, it already happened. If you were on the default model, you are on Fable 5.1 now, at High effort, and your cache-heavy sessions are already cheaper than they were in August.
The cases where it is worth actively checking:
- You pin a specific model by name. If your team's Claude Code config or gateway explicitly calls
claude-fable-5or an unconfiguredfable/bestalias, you are not automatically on 5.1. Update the pin to get the pricing and benchmark gains. - You run Claude Code through a corporate gateway. Ask whoever manages it whether Fable 5.1 has been enabled. Until it is, generic aliases keep resolving to the older Fable 5.
- You use GitHub Copilot alongside Claude Code. Fable 5.1 is also now generally available in GitHub Copilot, rolling out to Pro+, Max, Business, and Enterprise plans across VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, github.com, and mobile. One notable catch there: Fable 5.1 requires data retention by default in Copilot to power Anthropic's safety classifiers, unlike every other Claude model on the platform, which runs zero data retention. Eligible enterprises get a time-bound zero-data-retention exemption through year-end.
None of this requires you to relearn how you work. The interface, the tool-use patterns, the way you prompt Claude Code, none of it changed. What changed is what is happening underneath: a cheaper, more capable model doing the same job.
Claude Fable 5.1 for Product Managers
For a PM using Claude Code day to day, this default swap shows up less as a new feature and more as a quiet upgrade to work you already do. If you're also weighing this against other models on the market, our side-by-side comparison for product managers breaks down how Fable 5 stacks up against GPT-5.6 for PM workflows specifically. Longer agentic runs, competitive teardowns, multi-file PRD reviews, discovery synthesis across a stack of interview transcripts, all of that now runs on a model that Anthropic says is meaningfully stronger at root-cause analysis and multi-step follow-through, at a lower cache cost for exactly the kind of cache-heavy, read-the-same-context-repeatedly sessions those tasks produce.
It is also a reminder that the model under your agent changes more often than most teams track. If you built a workflow, a subagent, or a spec-review process around "whatever Claude Code defaults to," it is worth knowing that default just moved twice in three months.
Product manager and want to work like this? This is exactly what we teach in Claude Code for PMs, our live cohort for product teams: 3 live sessions of 90 minutes over 2 weeks. Every PM ships a real feature, builds their own agent, and gets personalized written feedback.
The Bottom Line
Claude Fable 5.1 replacing Sonnet 5 as Claude Code's default is not a headline feature launch, it is a quiet swap that lowers your bill on agentic sessions and raises the ceiling on what the model can carry through to completion. If you have not pinned a specific model, you are already benefiting from it. If you have, this is your cue to check whether it is time to update the pin.
Source: Claude Code changelog
