Top line
Claude Fable 5.2 is 20% cheaper per token, but some requests now fail
Anthropic launched Claude Fable 5.2 on Monday at 12:57 p.m. CT. Mythos 5.2, with the same capabilities, is limited to organizations verified through Anthropic's programs. Both cost $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for 5.1. Fast mode costs $16 and $80. Both have a 1M-token context window, and both require 30-day data retention, so zero data retention is unavailable unless Anthropic authorizes it. Three things now return a 400 error: sending a Fable 5.2 thinking block back after your code edits the system prompt, tools or an earlier message; declaring the old computer_20251124 tool (Bedrock still accepts it); and pairing a 5.2 executor with a non-5.2 advisor model. GitHub Copilot added Fable 5.2 for Pro+, Max, Business and Enterprise, but Business and Enterprise admins must turn on a policy that is off by default. Addy Osmani of Anthropic said that because the model "uses a lot of tokens," it costs "roughly the same as Fable 5.1" per task in their testing.
Why it matters: Test before you switch model IDs. Agents that rewrite their prompt mid-conversation or use the old computer-use tool will break, and the lower token price may not lower your bill.
Sources: Claude release notes · Anthropic announcement · GitHub Copilot changelog · Addy Osmani's post
Claude subscribers get Fable usage resets for a limited time
This is a launch offer, not a permanent limit change. Pro plans and standard Team seats get all Fable models inside their plan for one week, using up to 50% of weekly limits, plus one Fable reset. Both end Nov. 16 at 11:59 p.m. PT (1:59 a.m. CT Nov. 17). Max, Ultra, premium Team seats and premium seats on legacy seat-based Enterprise plans get one to three resets depending on how long they've paid, and those expire March 8, 2027 at 11:59 p.m. PT. Only accounts with paid billing before Nov. 9 at 10:00 a.m. PT (noon CT) qualify. Free, self-serve Enterprise, usage-based Enterprise and standard legacy Enterprise seats are excluded. A reset gives back Fable usage on the weekly limit and refills the five-hour window. It doesn't raise limits, and you trigger it yourself under Settings, then Usage.
Why it matters: Pro and Team users who want to try Fable have one week, and an unused reset is lost at the deadline.
Sources: Claude launch offers
Chrome fixes an exploited zero-day; CISA sets a 3-day deadline for four bugs
Chrome 157.0.8103.50/.51 fixes nine security bugs. Google says an exploit for CVE-2026-125350, a use-after-free in XML, "exists in the wild." Microsoft's Edge security notes say it is "actively working on releasing a security fix," so Edge isn't patched yet. Separately, CISA added four exploited bugs on Nov. 9, all due Nov. 12: OpenSSH CVE-2026-123158, a password-login bypass on builds using musl without PAM (Alpine), fixed in 10.6p2; GPUStack CVE-2026-102895, unauthenticated admin access in the AI model-serving cluster manager, fixed in 2.2.0; FreeSWITCH CVE-2026-116779; and IBM Aspera Faspex CVE-2026-123218.
Why it matters: Restart Chrome now and watch for Edge's update. Teams running Alpine containers with SSH password logins, or a GPUStack cluster that can be reached from the network, should patch today.
Sources: Chrome Releases · Edge security notes · CISA alert · OpenSSH release notes · GPUStack advisory
What to do
- Finance teams using OpenAI's Costs API: pull Nov. 6–9 exports again. OpenAI said exports published in that range had incorrect cost data; the incident was resolved at 1:02 a.m. CT, but the note doesn't say the records were fixed. OpenAI status
- EU and EFTA Microsoft 365 admins: decide on Claude before Dec. 9. Microsoft will turn Claude models on in Copilot for tenants that haven't made a choice, with processing in the EU Data Boundary. Message Center mirror
- Apple fleets: install iOS and macOS 27.2.1. Apple's Nov. 9 updates fix Bluetooth, AirPlay and iMessage bugs and the Brotli bug on CISA's list, which is due today; Apple doesn't say anything was exploited. Apple security releases
- Move off
gpt-image-2by Feb. 10, 2027. OpenAI deprecated it in favor ofgpt-image-2.6-sunburstorgpt-image-2.6-flare. OpenAI deprecations
Also notable
- OpenAI, Copilot, Cursor: evening errors. OpenAI's API had elevated errors from 5:08 to 5:50 p.m. CT, and Copilot rated GPT-5.3 Codex and GPT-5.6 a "major" incident from 5:16 to 6:05 p.m. CT. Cursor also saw errors on OpenAI models. OpenAI status · GitHub status
- GitHub: critical Actions and Checks API incident. It ran 12:12 to 12:59 p.m. CT, just after the pull_request_target brownout ended; GitHub hasn't linked the two. The next brownout is Nov. 16. GitHub status
- Meta: Muse Ember. Meta's smallest hosted model costs $0.10 per million input tokens and $0.50 per million output tokens, with a 1M-token context window; there are no open weights. Meta
- Meituan: LongCat-2.5 weights under MIT. The 1.6-trillion-parameter model (about 48 billion active) is now downloadable, a week after its API launch. Hugging Face
- Claude Code 2.1.321. It fixes MCP OAuth refresh tokens being sent to another host after a redirect, but it adds no secret scrubbing for the leak we covered Nov. 8. Release
- Catch-up: GitHub secret scanning for Actions logs. It's a public preview for Enterprise Cloud with Secret Protection or Advanced Security, off by default, and it doesn't scan old logs. GitHub changelog
Sources
Claude release notes · Anthropic announcement · GitHub Copilot changelog · Addy Osmani's post · Claude launch offers · Chrome Releases · Edge security notes · CISA alert · OpenSSH release notes · GPUStack advisory · OpenAI Costs API incident · Message Center mirror · Apple security releases · OpenAI deprecations · OpenAI API incident · Copilot incident · Actions incident · Meta · Hugging Face · Claude Code release · GitHub secret scanning changelog