September 1, 2026 was two announcements. In the morning, Anthropic made Claude Fable 5.1 generally available. Later the same day, OpenAI published “Path to Astra”, naming Astra as its next major model and saying it plans to make it available soon. Within hours, “Astra vs Fable 5.1” became a search, and it will be a genre by the end of the week.
Almost none of those pages can be honest, for a simple reason: one of the two models cannot be tested by anyone outside OpenAI. So this page does not pretend to a benchmark verdict. It compares what is actually knowable as of today, dated and sourced, in the same living format as our Fable 5.1 tracker and Astra tracker. The day Astra ships, this URL gets rewritten with the real comparison and a scorecard of what it got right.
On this page · 7 sectionsOpen
- September 1, 2026 was two announcements: Anthropic made Claude Fable 5.1 generally available, and OpenAI published ‘Path to Astra’, naming Astra as its next major model and saying it plans to make it available soon.
- Fable 5.1 is usable today: $10 per million input tokens and $50 per million output tokens, cache reads at $0.25 per million, API id claude-fable-5-1, a 1M-token context window. Astra has no date, no price and no API id.
- Astra’s rollout is gated on safety: OpenAI designated it the first model to meet the Critical cybersecurity threshold of its Preparedness Framework, and says it will reach a small alpha tester group first, then Daybreak Blue.
- No head-to-head benchmark exists and none can until Astra is testable. Fable 5.1’s published numbers are coding and knowledge work; Astra’s are cyber: ExploitBench 100%, two zero-days found in evaluation, 91.5% cyber-jailbreak refusal versus 59% for GPT-5.6 Sol, all OpenAI internal evaluations.
- Both vendors position their model for the same thing, long-horizon agent work. Fable 5.1 is the one you can run an agent on today; CellCog’s Agent Max and Team Max moved to it the day it shipped, with no change in pricing.
- This is a living page in the same format as our Fable 5.1 tracker: every fact dated and sourced, rewritten on the same URL the day Astra ships, with a scorecard of what this page got right.
- Is OpenAI Astra released?
- No. On September 1, 2026 OpenAI said ‘we plan to make Astra available soon’, starting with a small group of alpha testers. No date, price, tier or API identifier has been announced.
- Is Claude Fable 5.1 released?
- Yes. Anthropic made it generally available on September 1, 2026 at $10 and $50 per million tokens, the same headline price as Fable 5, with cache reads cut 75% to $0.25 per million.
- Which is better, Astra or Fable 5.1?
- Nobody outside OpenAI can say. Astra is not testable, and its published numbers are cybersecurity evaluations while Fable 5.1’s are coding and knowledge work. This page compares status, availability, pricing, safety posture and positioning, and will compare capability the day it becomes possible.
- What should I use today?
- For agent work you can ship now, Fable 5.1 is the frontier model that is actually available, and it is what CellCog’s Agent Max and Team Max tiers run on as of September 1. Astra is the one to watch, not the one to plan around.
§ 01The comparison, as of September 1, 2026
| Claude Fable 5.1 | OpenAI Astra | |
|---|---|---|
| Status | Generally available (Anthropic, September 1) | Announced, not released: “we plan to make Astra available soon” (OpenAI, September 1) |
| Release date | September 1, 2026 | None announced |
| Price | $10 / $50 per million input / output tokens; cache reads $0.25 per million | Not announced |
| API model id | claude-fable-5-1 | None |
| Who can use it | Anyone via Anthropic’s API and partners; CellCog Agent Max and Team Max | A small alpha tester group first, then Daybreak Blue; no general path |
| Safety posture | Production safeguards; Mythos 5.1 is the same model with permissive safeguards under restricted access | First model designated as meeting OpenAI’s Critical cybersecurity threshold; rollout gated on that |
| Published numbers | Terminal-Bench 4.0 55.8%, Terminal-Bench-Science 52.6%, CursorBench 3.2.0 73.4%, AutomationBench 31.4% (Anthropic) | ExploitBench 100%, two zero-days found in evaluation, 91.5% cyber-jailbreak refusal vs 59% for GPT-5.6 Sol (OpenAI internal evals) |
| Head-to-head | None exists | None exists |
| Positioning | Long-running coding and knowledge work; avoids shortcuts, fixes root causes | Long-horizon, multi-agent, persistent agents; math and cyber advances |
| What you can do today | Use it | Read about it |
§ 02What Fable 5.1 is
Anthropic’s most capable generally available model, released September 1 with a restricted sibling, Mythos 5.1. The positioning is long-horizon behavior: fewer easy-seeming shortcuts, root-cause fixes rather than symptom patches, progress updates while it works. The two facts that matter most for anyone running agents are economic. Anthropic held the headline price at Fable 5’s $10 and $50 per million tokens, and it cut cache reads by 75% to $0.25 per million. A standing agent re-reads its instructions, memory and conversation every turn, so that cut lands on exactly the kind of work agents do; Anthropic estimates savings of about 25% on typical workloads and up to about 45% on highly agentic ones.
The full spec record, the benchmark table, and our grading of the two-week rumor chain that preceded the release live on the Fable 5.1 tracker. The rest of the Claude 5.1 family, Opus and Sonnet, has its own tracker.
§ 03What Astra is, so far
OpenAI’s next major model, first named officially on August 1 when OpenAI credited “an internal version of Astra” with ten results on long-standing open problems in mathematics. On August 7, OpenAI said it could not rule out Astra meeting the Critical cybersecurity threshold of its Preparedness Framework and slowed development; through August a significant number of Astra workloads stayed paused.
The September 1 post resolves the question in the direction most people expected: Astra is designated the first model to meet that threshold. OpenAI describes the safeguards it built around that fact and reports three internal evaluation results, all cybersecurity: 100% on ExploitBench, two zero-day vulnerabilities discovered during evaluation, and a 91.5% refusal rate on cyber jailbreak attempts against 59% for GPT-5.6 Sol. It then says OpenAI plans to make Astra available soon, first to a small group of alpha testers and then to Daybreak Blue.
Three things the post does not contain: a date, a price, and an API identifier. Nor does it settle the name; OpenAI still simply calls it Astra, and the question of whether it ships as GPT-6 stays open. One availability fact is already official from a different post: OpenAI’s August 28 Cursor announcement says future models including Astra will not be supplied to Cursor after its acquisition by SpaceX. The full month-by-month record is on the Astra tracker.
§ 04Why there is no benchmark verdict
Two reasons, and they compound. Astra is not available to anyone who publishes benchmarks, so every number attached to it is OpenAI’s own. And the numbers OpenAI chose to publish are cybersecurity evaluations, because that is what the post is about; Anthropic’s are coding and knowledge work. There is no shared axis. A page that puts Fable 5.1’s 55.8% on Terminal-Bench 4.0 next to Astra’s 100% on ExploitBench and draws a conclusion is not comparing anything.
What is comparable is intent. Both vendors describe a model built for the same job: long-horizon, self-directed agent work rather than single-turn answers. Anthropic says it in product terms (ambitious, long-running projects); OpenAI says it in research terms (multiple agents collaborating on long-running tasks, training aimed at persistent agents). That convergence is the real September 1 story, and it is why the question that matters to a buyer is not which lab wins but which model you can put to work.
§ 05What each is for, and what to do today
If you are building or hiring agents now, Fable 5.1 is the frontier model that exists. CellCog’s Agent Max and Team Max tiers moved to it the afternoon it shipped, with no change in what you pay: a full shift of real work runs about $25, and the cost depends purely on how much work you assign. The day-one update has the details of what moved and why day one was safe.
If you are waiting for Astra, know what you are waiting for. The rollout OpenAI describes is alpha testers, then a named program, with no general-availability step announced. The exclusions are being decided now (Cursor already is one). And the model’s public identity so far is a cyber-capability designation, which shapes who gets it and under what monitoring. That is a model to watch closely, not a model to plan a quarter around.
Either way, the layer that decides how much a model change costs you is the one above the model. Memory, roles, permissions and org structure that live in the product rather than the model are what let an engine change on the day it ships. That was the CellCog experience with Fable 5.1 this afternoon, and it is the experience we intend when Astra arrives.
§ 06When this page flips
The day OpenAI makes Astra available, this URL is rewritten: the table above becomes a real specification comparison, we add the first independent benchmarks with their dates, and we grade this page’s own claims in a scorecard, the same way the Fable 5.1 tracker graded two weeks of rumors on release day. Until then, the triggers we are watching are a date, a price, an API identifier, the shipping name, and any third-party test of Astra’s capabilities outside cybersecurity.
§ 07Update log
September 1, 2026. Page created on the day of both announcements. Facts drawn from Anthropic’s Claude Fable and Mythos 5.1 release page and model documentation, and from OpenAI’s “Path to Astra” post, both retrieved September 1.
Q1What did OpenAI announce on September 1, 2026?
A post titled ‘Path to Astra: critical capabilities and frontier safeguards’. It names Astra as OpenAI’s next major model, designates it the first model to meet the Critical cybersecurity threshold of the Preparedness Framework, describes the safeguards built for that, reports internal evaluation results (ExploitBench 100%, two zero-day vulnerabilities found during evaluation, 91.5% refusal on cyber jailbreak attempts versus 59% for GPT-5.6 Sol), and says OpenAI plans to make Astra available soon, first to a small alpha tester group and then to Daybreak Blue. It gives no date, price or API identifier.
Q2What did Anthropic announce on September 1, 2026?
General availability of Claude Fable 5.1 alongside Mythos 5.1, a restricted variant of the same model with permissive safeguards for vetted cyberdefense and life-science organizations. Fable 5.1 keeps Fable 5’s $10 and $50 per million token pricing, cuts cache reads 75% to $0.25 per million, and reports 55.8% on Terminal-Bench 4.0, 52.6% on Terminal-Bench-Science 0.1, 73.4% on CursorBench 3.2.0 and 31.4% on AutomationBench, all Anthropic’s own numbers. Our full record is on the Fable 5.1 tracker.
Q3How much will Astra cost compared with Fable 5.1?
Unknown. Fable 5.1 is $10 per million input tokens and $50 per million output tokens. OpenAI has announced no Astra pricing; the only number near it is OpenAI’s August 1 note that the tokens behind Astra’s ten mathematical results would cost roughly $2,000 at GPT-5.6 Sol rates, which is a research-compute estimate, not a price.
Q4Is Astra GPT-6?
Undecided as far as anyone outside OpenAI knows. The Information reported in July that OpenAI had not settled whether Astra ships as GPT-6, a GPT-5-series release, or under another name, and the September 1 post simply calls it Astra. ‘GPT-6 Astra’ is a search phrase, not a product.
Q5What does OpenAI's Critical cybersecurity designation mean?
Per OpenAI’s August 7 definition, a model reaches Critical if it can identify and develop working zero-day exploits across many hardened real-world systems without human intervention, or devise and execute a novel end-to-end cyberattack against hardened targets from a high-level goal. On September 1 OpenAI stated Astra is the first model designated as meeting that threshold, which is why its rollout is gated rather than general.
Q6Can I run Fable 5.1 on CellCog?
Yes. CellCog’s Agent Max and Team Max tiers moved to Fable 5.1 on September 1, 2026, the day it shipped. Your pricing does not change: a full shift of real work runs about $25, and the cost depends purely on how much work you assign. When Astra becomes available, the engine can change under your AI employee without your setup changing.