1,050,000 Tokens: GPT-6 Astra vs Claude Fable 5.1 in 3 Numbers

Two frontier models landed in the same week. Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on 1 September 2026. OpenAI released GPT-6 Astra on 3 September 2026. Both companies called their own release the best in the world — Anthropic said "the world's most advanced models for coding and knowledge work", OpenAI said "the world's most intelligent and aligned model".

Marketing aside, the specifications are public and they are close enough that the choice comes down to how you actually work. Here is what is verifiable.

1. The sticker price is identical — $10 in, $50 out

Both flagships charge the same headline rate: $10 per million input tokens and $50 per million output tokens. That is unusual, and it means a per-token comparison tells you almost nothing on its own.

Both offer a 50% discount for batch processing. Astra adds a Fast mode at 2x the price for up to 2x the speed, and charges more for very long prompts: requests over 272,000 input tokens are billed at 2x input and 1.5x output for the whole request, not just the excess.

2. Cached context is where the money actually goes

Agentic work re-reads the same things constantly — the same repository, the same system instructions, the same tool definitions, the same growing conversation. Those re-reads are billed as cache reads, and on long-running work they are most of the bill.

This is the one number worth sitting with. Fable 5.1's cached input is a quarter of Astra's, and only 25% more than Sonnet 5's — despite its base input price being five times Sonnet's. It is a deliberate bet that the future of the bill is agents re-reading context, not fresh prompts.

3. Effort settings turned into a price dial

Both models now expose several reasoning-effort levels, and the gap between the cheapest and the most expensive setting is large enough that the setting matters more than the model choice for many jobs.

• GPT-6 Astra: low, medium, high, xhigh, max. The API default is low — the marketing highlights the top end, so set this explicitly if you want it.
• Claude Fable 5.1: five effort levels, defaulting to High in Claude Code and Medium in Claude's apps.

What each one is actually built for

Reading past the superlatives, the two launches emphasise different things.

OpenAI pushed computer use hardest: filling forms, driving spreadsheets, building sites, operating a browser end to end. It is also the first OpenAI model designated as meeting the company's critical cybersecurity capability threshold, which is why it rolled out to vetted cybersecurity customers first rather than to everyone.

Anthropic's numbers lean towards long, autonomous work. On its own published table, Fable 5.1 scores 52.6% on Terminal-Bench-Science 0.1 against Fable 5's 24.7% — more than double — and 31.4% on AutomationBench against 17.1%. The short-horizon benchmarks moved only a few points. The longer the task runs, the bigger the gap.

What this means if you are paying in rupees

Neither of these is the model most people should be sending everyday traffic to. Both are the top of a range, priced accordingly, and both vendors say so themselves — Anthropic's own documentation tells developers to start with Opus 5 and reach for Fable 5.1 only when Opus at higher effort still falls short.

Mythos 5.1: the same model behind a door

Worth knowing because it explains the odd benchmark rows: Claude Mythos 5.1 is the same model as Fable 5.1, with identical weights and pricing, running with more permissive safeguards in biology and cybersecurity. It is not a consumer product and not generally available — access runs through invitation-only verification programmes, currently limited to US organisations.