--- title: GPT-6 Astra Review: Benchmarks, Pricing, and the Catch description: GPT-6 Astra, OpenAI's new flagship model (gpt-6-astra), launched September 3 with the same headline pricing as Claude Fable 5.1 ($10/$50 per million tokens)... canonical: https://daily.dev/posts/gpt-6-astra-review-benchmarks-pricing-and-the-catch-kajxyhzam twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: GPT-6 Astra Review: Benchmarks, Pricing, and the Catch | daily.dev og:description: GPT-6 Astra, OpenAI's new flagship model (gpt-6-astra), launched September 3 with the same headline pricing as Claude Fable 5.1 ($10/$50 per million tokens)... og:url: https://daily.dev/posts/gpt-6-astra-review-benchmarks-pricing-and-the-catch-kajxyhzam og:image: https://api.daily.dev/og/posts/KAJXyhzaM.png og:image:alt: GPT-6 Astra Review: Benchmarks, Pricing, and the Catch og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # GPT-6 Astra Review: Benchmarks, Pricing, and the Catch **[Alex CloudStar](https://daily.dev/sources/alexcloudstar)** · [@alexcloudstar](https://daily.dev/alexcloudstar) · 16 min read · 0 upvotes · 0 comments ## Summary GPT-6 Astra, OpenAI's new flagship model (gpt-6-astra), launched September 3 with the same headline pricing as Claude Fable 5.1 ($10/$50 per million tokens) but cache reads that are four times more expensive ($1.00 vs $0.25 per million), a critical detail for agentic workloads that repeatedly re-read context. Astra posts a large jump on Terminal-Bench 4.0 (57.9% vs 37.3%) and scores 100% on ExploitBench, making it the first model OpenAI classifies at the Critical cybersecurity tier, with exploit generation refused publicly and fuller access gated behind Trusted Access and the Daybreak Blue program. It is not a clean sweep: it trails Fable 5.1 on Humanity's Last Exam with tools and shows only marginal gains on DeepSWE. Notably, it uses a 'recurrent depth' reasoning technique that loops computation through latent space, which OpenAI's own system card says reduces chain-of-thought monitorability and can let the model evade some monitors on adversarial tasks. Other changes include Codex 'notes' replacing lossy context summarization, a default low reasoning effort that undersells marketed benchmarks unless explicitly raised, no fine-tuning support, and a pricing cliff that doubles input/cache costs past 272K tokens. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much does GPT-6 Astra cost per million tokens compared to Claude Fable 5.1? GPT-6 Astra and Claude Fable 5.1 have identical base prices at $10 per million input tokens and $50 per million output tokens. The real difference is cache reads: Astra charges $1.00 per million while Fable 5.1 charges $0.25 per million, a fourfold gap that matters most for agentic workloads re-reading the same context on every loop turn. Astra also doubles input and cache pricing past 272,000 tokens in a single prompt. _Developers pricing out agent workloads across models can track pricing shifts like this via daily.dev._ ### What reasoning effort does the GPT-6 Astra API use by default? The GPT-6 Astra API defaults to low reasoning effort, not the high or max settings OpenAI used to produce its marketed benchmark scores. Developers must explicitly set the reasoning.effort parameter to low, medium, high, xhigh, or max to reach the performance shown in launch benchmarks, otherwise they are running a materially weaker configuration than advertised. _Anyone evaluating a new model release can catch defaults like this early by following model updates on daily.dev._ ### Why is GPT-6 Astra's cybersecurity capability restricted behind a verification program? GPT-6 Astra is the first OpenAI model classified at the Critical tier of its Preparedness Framework, meaning it can find and exploit novel vulnerabilities in hardened targets without step-by-step human guidance, evidenced by a 100% ExploitBench score and reported discovery of genuine zero-days during evaluation. The public model refuses exploit generation, while less restricted access requires Trusted Access or the Daybreak Blue defender program. _Security tooling builders navigating access tiers like this can stay current on AI model capabilities via daily.dev._ ## Similar posts on daily.dev - [GPT-6 Astra: The AI That Doesn’t Just Answer. It Gets the Job Done](https://daily.dev/posts/gpt-6-astra-the-ai-that-doesn-t-just-answer-it-gets-the-job-done-zotuejbng) · Medium · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#openai](https://daily.dev/tags/openai), [#ai-safety](https://daily.dev/tags/ai-safety) [View this post on daily.dev](https://daily.dev/posts/gpt-6-astra-review-benchmarks-pricing-and-the-catch-kajxyhzam) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"GPT-6 Astra Review: Benchmarks, Pricing, and the Catch","url":"https://daily.dev/posts/gpt-6-astra-review-benchmarks-pricing-and-the-catch-kajxyhzam","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/gpt-6-astra-review-benchmarks-pricing-and-the-catch-kajxyhzam"},"datePublished":"2026-09-05T21:09:52.400Z","dateModified":"2026-09-06T18:00:17.441Z","description":"GPT-6 Astra, OpenAI's new flagship model (gpt-6-astra), launched September 3 with the same headline pricing as Claude Fable 5.1 ($10/$50 per million tokens)...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/fdf1792a9b9c964c29b8c7d0bec2e8d2?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/fdf1792a9b9c964c29b8c7d0bec2e8d2?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Alex CloudStar","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Person","name":"Alex Cloudstar","url":"https://daily.dev/alexcloudstar","image":"https://media.daily.dev/image/upload/s--xA-1aKK7--/f_auto/v1767564650/avatars/avatar_ZUjetzkwG7V5behQOpH0X?_a=BAMAK+ZW0","description":"Hi there! 👋🏻 I'm a Senior Full-Stack Developer oriented on Javascript","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":990}},"commentCount":0,"discussionUrl":"https://daily.dev/posts/gpt-6-astra-review-benchmarks-pricing-and-the-catch-kajxyhzam","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,openai,ai-safety","timeRequired":"PT16M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Alex CloudStar","item":"https://daily.dev/sources/alexcloudstar"},{"@type":"ListItem","position":3,"name":"GPT-6 Astra Review: Benchmarks, Pricing, and the Catch"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/gpt-6-astra-review-benchmarks-pricing-and-the-catch-kajxyhzam#faq","mainEntity":[{"@type":"Question","name":"How much does GPT-6 Astra cost per million tokens compared to Claude Fable 5.1?","acceptedAnswer":{"@type":"Answer","text":"GPT-6 Astra and Claude Fable 5.1 have identical base prices at $10 per million input tokens and $50 per million output tokens. The real difference is cache reads: Astra charges $1.00 per million while Fable 5.1 charges $0.25 per million, a fourfold gap that matters most for agentic workloads re-reading the same context on every loop turn. Astra also doubles input and cache pricing past 272,000 tokens in a single prompt. Developers pricing out agent workloads across models can track pricing shifts like this via daily.dev."}},{"@type":"Question","name":"What reasoning effort does the GPT-6 Astra API use by default?","acceptedAnswer":{"@type":"Answer","text":"The GPT-6 Astra API defaults to low reasoning effort, not the high or max settings OpenAI used to produce its marketed benchmark scores. Developers must explicitly set the reasoning.effort parameter to low, medium, high, xhigh, or max to reach the performance shown in launch benchmarks, otherwise they are running a materially weaker configuration than advertised. Anyone evaluating a new model release can catch defaults like this early by following model updates on daily.dev."}},{"@type":"Question","name":"Why is GPT-6 Astra's cybersecurity capability restricted behind a verification program?","acceptedAnswer":{"@type":"Answer","text":"GPT-6 Astra is the first OpenAI model classified at the Critical tier of its Preparedness Framework, meaning it can find and exploit novel vulnerabilities in hardened targets without step-by-step human guidance, evidenced by a 100% ExploitBench score and reported discovery of genuine zero-days during evaluation. The public model refuses exploit generation, while less restricted access requires Trusted Access or the Daybreak Blue defender program. Security tooling builders navigating access tiers like this can stay current on AI model capabilities via daily.dev."}}]} ```