OpenAI and Anthropic shipped new flagship models a week apart. Here is what changed on a $20 plan

Two of the three big model makers released their top models within 48 hours of each other, and for once the interesting part is not who scored higher. Both announcements describe the same shift: a model that does not just write the email but opens the software, fills the form, and updates the record.

What OpenAI says Astra does

OpenAI's own description of GPT-6 Astra leads with computer use. The model, it says, "can take care of tedious tasks like filling out online forms, updating customer records in a CRM, and organizing your calendar", and can "conduct online research and draft summaries in your email or in your document editor." It is also described as better at handling vague instructions: it "uses context to fill in routine gaps and asks focused questions when the answer could change the outcome."

On cost, OpenAI says Astra is trained "to complete tasks in fewer tokens with fewer retries, which means less rework and lower cost per task." For a subscriber that is invisible; for anyone paying by the token through the API, it is the number to watch on next month's bill.

One line in the availability section matters for business accounts: "Enterprise administrators can enable Astra for their workspace; access is off by default at launch." If your team uses a Business or Enterprise plan and nobody has flipped that switch, you do not have it yet.

What Anthropic says Fable 5.1 does

Anthropic's announcement is organized around three customer complaints it says it heard: price, data retention, and safeguards. On price, Fable 5.1 "will cost an estimated 25% less than Fable 5 for typical workloads, wherever usage is billed by token," because the company cut the price of cache reads by 75%, to $0.25 per million tokens. The headline input and output prices did not change.

On the consumer side the change is quieter. The post notes that Fable 5.1 "defaults to High effort in Claude Code, and to Medium in Claude Cowork and on Claude.ai", meaning the version most subscribers get is tuned for speed and cost rather than maximum capability. Anthropic says that at Low or Medium effort the new model matches or beats the old one at a lower cost.

The same release introduced Claude Mythos 5.1, which Anthropic describes as the same model "with different levels of safeguards," available only to vetted cybersecurity and life-science organizations through a US government-coordinated access program. That one is not for you, and it is not meant to be.

Both models: $10 per million input tokens, $50 per million output tokens on their APIs. Both included in existing paid subscriptions. Neither company raised the price of the plan you already have.

The safety claims, and who is making them

Both companies make strong safety statements, and both are describing their own tests. OpenAI says that on its internal computer-use safety benchmark, covering scenarios "such as exposing confidential information, sharing a dashboard too broadly, or deleting data," Astra "produced unintended outcomes 89% less often than GPT-5.6 Sol and 74.7% less often than Claude Fable 5.1." That is OpenAI measuring a competitor on OpenAI's test. Anthropic, for its part, says its new cybersecurity safeguards "block 60% fewer false positives than before" and that eligible customers can run Fable 5.1 with zero data retention.

What this means at your desk

For a business on a $20 plan, the practical change is that the model behind your saved prompts just got swapped. That is usually good and occasionally not: a model that is better at inferring intent can also infer a different intent than the one you had. This is exactly the case OpenAI's own prompting guide flags when it recommends pinning "specific model snapshots" to "ensure consistent behavior". Re-run your most important prompt once and read the output.

The computer-use features are the ones to approach slowly. A model that can update your CRM can also update it wrong, and OpenAI's decision to ship Astra off by default for enterprise workspaces tells you how the company itself is thinking about that. Start with something you can undo.