Model comparison
Claude Fable 5.1 vs GPT-6 Astra: why both cost $10.50
Claude Fable 5.1 and GPT-6 Astra share $10 input and $50 output prices, and a cached coding session costs $10.50 on each. Caching decides which way it tips.
· Prices as of September 28, 2026
Claude Fable 5.1
Anthropic · Released September 1, 2026
Anthropic's most capable generally available model, aimed at demanding reasoning and long-horizon agentic coding. Anthropic suggests it when Opus-tier results fall short.
Claude Fable 5.1 facts and comparisonsGPT-6 Astra
OpenAI · Released September 3, 2026
OpenAI's top GPT-6 model and its most expensive, aimed at the hardest long-running work that spans many tools, including coding.
GPT-6 Astra facts and comparisons
The short answer
Claude Fable 5.1 and GPT-6 Astra have identical list prices, and the example agentic coding session costs $10.50 on each. Fable 5.1 charges $0.25 per million for a cache hit against $1 on Astra, but its 1-hour cache writes cost $20 against Astra's $12.50, and in this session the two differences cancel out. Pick Fable 5.1 for Claude Code and for prompts beyond 272K input tokens, which it bills at standard rates, and GPT-6 Astra for Codex, whose CLI ships it as the default model.
Choose Claude Fable 5.1 if
- You work in Claude Code, where /model fable selects it, or reach it through Cursor, OpenCode, OpenRouter, or GitHub Copilot.
- Your requests often pass 272K input tokens: Fable 5.1 bills its whole 1M window at standard rates, while GPT-6 Astra charges 2x for input and cache and 1.5x for output above that point.
- Your sessions reread the same cached context many times, since each Fable 5.1 hit costs 2.5% of its input price.
- Anthropic's own suggestion describes your situation: Opus-tier results fall short on your hardest tasks.
Choose GPT-6 Astra if
- You work in Codex, where the CLI's bundled model list (version 0.158.0) makes GPT-6 Astra the default at low reasoning effort.
- Your caches are written often and read only a few times before they change, since every Astra write costs $12.50 per million while a Fable 5.1 1-hour write costs $20.
- You want to check OpenAI's claim that Astra reaches stronger results with substantially fewer output tokens, which OpenAI says lowers the estimated cost per task.
Side by side
Specs and prices
| Fact | Claude Fable 5.1 | GPT-6 Astra |
|---|---|---|
| Maker | Anthropic | OpenAI |
| API model id | claude-fable-5-1 | gpt-6-astra |
| Released | September 1, 2026 | September 3, 2026 |
| Status | Current | Current |
| Context window | 1M tokens | 1.05M tokens |
| Max output | 128K tokens | 128K tokens |
| Open weights | No | No |
| Input, per 1M tokens | $10 | $10 |
| Cache hit, per 1M | $0.25 | $1 |
| Cache write, per 1M | $12.50 (5-minute), $20 (1-hour) | $12.50 |
| Output, per 1M tokens | $50 | $50 |
| Runs in | Claude Code, Cursor, OpenCode, OpenRouter, and GitHub Copilot | Codex, OpenCode, OpenRouter, and GitHub Copilot |
Standard API rates in US dollars, as published by each maker (Claude Fable 5.1: September 26, 2026; GPT-6 Astra: September 28, 2026). Batch and priority tiers, taxes, and subscription plans are not included. Claude Fable 5.1: The full 1M context window is billed at standard rates. GPT-6 Astra: Requests over 272K input tokens cost 2x for input and cache and 1.5x for output, for the whole request.
Cost
What the same work costs
The same token counts, priced at each model’s published rates. Cached tokens are billed at each maker’s cache prices, so the session shows what caching is worth on each model.
Real sessions differ: the two models count the same code as different numbers of tokens, and reasoning settings change how much each one writes. Your own history is the real test.
| Workload | Claude Fable 5.1 | GPT-6 Astra |
|---|---|---|
| Agentic coding session, 100K input, 400K written to cache, 2M read from cache, 50K output | $10.50 | $10.50 |
| Large one-off review, 150K input with no cache hits, 10K output | $2.00 | $2.00 |
| Output-heavy generation, 30K input, 80K output | $4.30 | $4.30 |
| A month of sessions, 110 sessions: 5 a day, 22 working days | $1,155.00 | $1,155.00 |
| Where the session’s cost goes | ||
| Cache writes | $6.50 | $5.00 |
| Cache reads | $0.50 | $2.00 |
| Uncached input | $1.00 | $1.00 |
| Output | $2.50 | $2.50 |
- caching saves on the session with Claude Fable 5.1 (62%)
- $17.00
- caching saves on the session with GPT-6 Astra (62%)
- $17.00
Why the example session costs $10.50 on both
Claude Fable 5.1 and GPT-6 Astra publish the same headline rates: $10 per million input tokens, $50 per million output tokens, and $12.50 for a standard cache write. That write price is 1.25x input at both makers, Anthropic's 5-minute write and OpenAI's single write from GPT-5.6 on. With no caching involved, the two cost the same to the cent: $2.00 for the large one-off review and $4.30 for the output-heavy generation.
The agentic coding session is where the caching rules diverge. Anthropic prices a Fable 5.1 cache hit at 0.025x input, or $0.25 per million, while OpenAI charges 0.1x, or $1. Anthropic also sells a 1-hour cache write at 2x input, $20 per million, where OpenAI has a single write price. The session writes 200K tokens to each of Anthropic's two caches and reads 2M tokens back.
Add it up and the two differences cancel. Cache writes cost $6.50 on Fable 5.1 and $5.00 on Astra, a $1.50 gap in Astra's favor. Cache reads cost $0.50 on Fable 5.1 and $2.00 on Astra, a $1.50 gap the other way. Input and output match, so both sessions land on $10.50, and a month of 110 sessions on $1,155.00.
Which way does the tie break?
The tie is a property of this particular mix of writes and reads, not a rule. Every extra read of a cached prefix makes Fable 5.1 relatively cheaper, because each hit costs a quarter of Astra's. Every extra 1-hour write makes Astra relatively cheaper. Anthropic's rule of thumb is that a 1-hour write pays for itself after two cache reads, and a 5-minute write after one.
Claude Code shows how much the setting matters. Its main conversation uses the 1-hour cache on a Claude subscription and the 5-minute cache with an API key. With an API key, Fable 5.1 writes at $12.50 per million, the same as Astra, and the cheaper reads then tip the session toward Fable 5.1. OpenAI has one write price and no separate 1-hour option; on GPT-5.6 and later models a cached prefix stays reusable for at least 30 minutes after its last use.
Caching saves the same amount on both in this example, $17.00, or 62% of what the same tokens would cost uncached. The cache math on the homepage walks through the same session line by line.
What happens above 272K input tokens?
The largest price difference between these models is not in the cost table. OpenAI bills any GPT-6 Astra request over 272K input tokens at 2x for input and cache and 1.5x for output, for the whole request, and Astra accepts at most 922K input tokens of its 1.05M window. Anthropic bills Fable 5.1's full 1M context window at standard rates.
Every request in the example session stays under 200K input tokens, so neither rule shows up there. If your agent loads large parts of a repository into a single prompt, though, Astra's long-context rate applies to the whole request, output included, while Fable 5.1's rates stay flat. Both models write up to 128K tokens per response.
How Anthropic and OpenAI pitch their top models
Each sits at the top of its maker's current line. Anthropic calls Fable 5.1 and its restricted twin, Claude Mythos 5.1, "the world's most advanced models for coding and knowledge work" and aims Fable 5.1 at demanding reasoning and long-horizon agentic coding. It suggests Fable 5.1 when Opus-tier results, such as those from Claude Opus 5.5, fall short.
OpenAI describes GPT-6 Astra as "Our most capable model, built for the hardest end-to-end work" and says it stays coherent during long tasks better than GPT-5.6 Sol and earlier models. It also says Astra reached stronger results with substantially fewer output tokens in several evaluations. The cost table prices the same output token count on both models, so it cannot show that effect.
Defaults push in opposite directions. Fable 5.1 runs adaptive thinking at high effort by default, while Codex starts Astra at low reasoning effort, and effort changes how many tokens each one writes. Tokenizers differ too: Fable 5.1 uses Anthropic's newer tokenizer, which counts about 30% more tokens than earlier Claude models for the same text. Price a week of your own work on both before settling on one.
Prompt caching
How each maker bills cached tokens
Anthropic
Claude caches a prompt prefix up to a breakpoint. One top-level cache_control field places the breakpoint automatically and moves it as the conversation grows, or you can mark up to 4 blocks yourself. Claude Code manages caching for you.
A 5-minute cache write costs 1.25x the input price and a 1-hour write costs 2x. A cache hit costs 0.1x input on most models, 0.05x on Claude Opus 5.5, and 0.025x on Claude Fable 5.1 and Claude Mythos 5.1. Every hit restarts the cache lifetime at no charge.
Anthropic's rule of thumb: a 5-minute write pays for itself after one cache read, and a 1-hour write after two.
In Claude Code, the main conversation uses the 1-hour cache on a Claude subscription and the 5-minute cache with an API key. Each model has its own cache, so switching models starts over.
Source: Anthropic: Prompt caching
OpenAI
Prompt caching is on by default. From GPT-5.6 on you can also mark up to four explicit cache breakpoints, while GPT-5.5 and earlier cache automatically only.
From GPT-5.6 on, a cache write costs 1.25x the uncached input price and a cache hit costs 0.1x. GPT-5.5 and earlier add no charge for writing the cache: written tokens are billed as ordinary input, and a hit costs 0.1x on the models compared here.
On GPT-5.6 and later, a cached prefix stays reusable for at least 30 minutes after its last use, and caching starts at 1,024 input tokens.
Source: OpenAI: Prompt caching
Your own numbers
See what Claude Fable 5.1 and GPT-6 Astra really cost you.
everyaitoken reads your Claude Code, Cursor, OpenCode, OpenRouter, and Codex history on your Mac and prices every request at API rates, with what caching saved or cost. $9 once.
FAQ
Questions
Is Claude Fable 5.1 cheaper than GPT-6 Astra?
At list prices, neither is cheaper: both charge $10 per million input tokens and $50 per million output tokens. All three example workloads cost the same on each, and 110 agentic sessions come to $1,155.00 a month on either. Fable 5.1 pulls ahead when a session rereads its cache many times, and Astra pulls ahead when caches are written often and read little.
Which coding tools use each model by default?
Codex CLI's bundled model list, version 0.158.0, makes GPT-6 Astra its default. Astra also runs in the Codex app and IDE extension, but not in Codex cloud. Fable 5.1 is not the default on any Claude Code plan, so you select it with /model fable.
Can Claude Fable 5.1 run under zero data retention?
No. Anthropic requires 30-day data retention for Fable 5.1, so it is not available under zero data retention. The sources behind this page do not cover GPT-6 Astra's retention terms, so check those with OpenAI.
How can I see what each model costs on my own work?
EveryToken is a $9 one-time Mac app that reads your local Claude Code and Codex history, prices every request at the makers' API rates, and shows what caching saved or cost. Its figures are API-equivalent estimates at published rates, and a Claude or ChatGPT subscription is priced differently.
Sources
- Anthropic: Pricing
- Anthropic docs: Claude Fable 5.1
- Anthropic: Claude Fable 5.1 and Claude Mythos 5.1
- Claude Code docs: Model configuration
- Cursor docs: Claude Fable 5.1
- OpenRouter: Claude Fable 5.1
- OpenCode docs: Zen
- GitHub Docs: Supported AI models in Copilot
- OpenAI: API pricing
- OpenAI docs: GPT-6 Astra
- OpenAI: Using the latest model
- OpenAI: API changelog
- Codex docs: Models
- Codex CLI: bundled model catalog
- OpenRouter: GPT-6 Astra
- Anthropic: Prompt caching
- OpenAI: Prompt caching