Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, the newest additions to its flagship model line aimed at software engineering and long-horizon knowledge work. Fable 5.1 is available broadly through the Claude API, while Mythos 5.1 stays restricted to vetted participants in Anthropic's Project Glasswing program. Both models handle up to one million tokens of context with output reaching 128,000 tokens per response, and API pricing holds at $10 per million input tokens and $50 per million output, with cache read costs reduced.
The update targets developers and research teams running long agentic sessions of code generation, debugging, and document analysis. Fable 5.1 adds mid-conversation effort control, letting users dial how hard the model works on a given step without starting a new session, and it posts leading scores on coding evaluations including SWE-bench and Terminal-Bench 2.1. The timing matters because 5.1 arrives roughly three months after the Fable 5 and Mythos 5 generation shipped in June 2026, marking a faster cadence as Anthropic pushes against GPT-5.6 and other frontier systems on coding and reasoning workloads.
Anthropic was founded in 2021 by Dario Amodei and Daniela Amodei, along with several former OpenAI researchers including Jared Kaplan, Jack Clark, Chris Olah, Ben Mann, Sam McCandlish, and Tom Brown. The company operates as a public benefit corporation focused on AI systems that are steerable, interpretable, and safe, and this release reflects that stance by gating the more capable Mythos 5.1 behind a controlled access program while shipping Fable 5.1 to the broader developer base.
Comments (13)
Sign in to join the discussion.
P
Priya NadkarniSep 1, 2026
Fable and Mythos is such a lore-pilled naming choice, I already know the sonnets are next. Betting the enterprise crowd renames their internal wikis by Friday.
R
Rin TakedaSep 1, 2026
Great, now do a smaller one for latency-sensitive stuff and a voice mode and a canvas and honestly why not a browser too.
O
Ole K.Sep 1, 2026
The launch page loads in under a second and the demo has zero cursor jitter, which tells me someone at Anthropic actually respects the viewer. Meanwhile most model launches ship a 4K video of someone squinting at a terminal.
M
Mattes BergqvistSep 1, 2026
We had a prototype internally in 2019 that did structured knowledge retrieval with human-in-the-loop grading. Different scale obviously, but the shape of the problem hasn't moved as much as people think.
K
Kojo A.Sep 1, 2026
Curious how many people it actually took to ship this vs the perception. Solo devs are eating well downstream either way.
R
renataSep 1, 2026
Every launch tweet from a frontier lab now reads like the trailer for a Denis Villeneuve movie. Where is the changelog, where are the eval numbers, why am I being told a story.
Y
Yusuf DemirSep 1, 2026
Two SKUs at the frontier tier is a pricing sommelier move. Somebody in finance is about to discover their team is quietly on the more expensive one.
A
Amaka O.Sep 1, 2026
Wait, what's the actual difference between Fable and Mythos though? The tweet says both are for coding and knowledge work, so which one do I tell my team to use.
D
Dhruv IyerSep 1, 2026
World's most advanced at coding and knowledge work is basically saying world's most advanced at what my job used to be. Cool cool cool.
L
Lin C.Sep 1, 2026
Would love to know if that agentic coding demo was one take or lightly edited. The pause before the tool call felt suspiciously polite.
S
Sami HaddadSep 1, 2026
Every benchmark chart in these announcements has the y-axis start at 62%. I'll believe the coding claims when it survives one week of my legacy Django repo.
C
Carmela RicciSep 1, 2026
Before anyone in my Slack asks: no, you cannot pipe your customer PII through Mythos just because the name sounds impressive. Please wait for the data residency doc.
B
Benno FischerSep 1, 2026
One of my portfolio companies rewrote their whole eval harness overnight to bench this. If your product roadmap moves based on a Tuesday tweet, you probably weren't building a moat.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Anthropic
Al research and products that put safety at the frontier
Al research and products that put safety at the frontier
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, the newest additions to its flagship model line aimed at software engineering and long-horizon knowledge work. Fable 5.1 is available broadly through the Claude API, while Mythos 5.1 stays restricted to vetted participants in Anthropic's Project Glasswing program. Both models handle up to one million tokens of context with output reaching 128,000 tokens per response, and API pricing holds at $10 per million input tokens and $50 per million output, with cache read costs reduced. The update targets developers and research teams running long agentic sessions of code generation, debugging, and document analysis. Fable 5.1 adds mid-conversation effort control, letting users dial how hard the model works on a given step without starting a new session, and it posts leading scores on coding evaluations including SWE-bench and Terminal-Bench 2.1. The timing matters because 5.1 arrives roughly three months after the Fable 5 and Mythos 5 generation shipped in June 2026, marking a faster cadence as Anthropic pushes against GPT-5.6 and other frontier systems on coding and reasoning workloads. Anthropic was founded in 2021 by Dario Amodei and Daniela Amodei, along with several former OpenAI researchers including Jared Kaplan, Jack Clark, Chris Olah, Ben Mann, Sam McCandlish, and Tom Brown. The company operates as a public benefit corporation focused on AI systems that are steerable, interpretable, and safe, and this release reflects that stance by gating the more capable Mythos 5.1 behind a controlled access program while shipping Fable 5.1 to the broader developer base.
Comments (13)
Fable and Mythos is such a lore-pilled naming choice, I already know the sonnets are next. Betting the enterprise crowd renames their internal wikis by Friday.
Great, now do a smaller one for latency-sensitive stuff and a voice mode and a canvas and honestly why not a browser too.
The launch page loads in under a second and the demo has zero cursor jitter, which tells me someone at Anthropic actually respects the viewer. Meanwhile most model launches ship a 4K video of someone squinting at a terminal.
We had a prototype internally in 2019 that did structured knowledge retrieval with human-in-the-loop grading. Different scale obviously, but the shape of the problem hasn't moved as much as people think.
Curious how many people it actually took to ship this vs the perception. Solo devs are eating well downstream either way.
Every launch tweet from a frontier lab now reads like the trailer for a Denis Villeneuve movie. Where is the changelog, where are the eval numbers, why am I being told a story.
Two SKUs at the frontier tier is a pricing sommelier move. Somebody in finance is about to discover their team is quietly on the more expensive one.
Wait, what's the actual difference between Fable and Mythos though? The tweet says both are for coding and knowledge work, so which one do I tell my team to use.
World's most advanced at coding and knowledge work is basically saying world's most advanced at what my job used to be. Cool cool cool.
Would love to know if that agentic coding demo was one take or lightly edited. The pause before the tool call felt suspiciously polite.
Every benchmark chart in these announcements has the y-axis start at 62%. I'll believe the coding claims when it survives one week of my legacy Django repo.
Before anyone in my Slack asks: no, you cannot pipe your customer PII through Mythos just because the name sounds impressive. Please wait for the data residency doc.
One of my portfolio companies rewrote their whole eval harness overnight to bench this. If your product roadmap moves based on a Tuesday tweet, you probably weren't building a moat.