Claude news roundup: July to October 2026

This is a dated summary of what changed with Claude between 1 July and early October 2026. Every fact comes from Anthropic’s own pages: the news posts, the Claude Platform release notes, the models overview and the Claude Code changelog. Benchmark numbers are Anthropic’s, run by Anthropic or its named partners, and I have not reproduced any of them. The last section is my opinion and is marked as such.

Timeline

Date	What happened
1 July	Access to Claude Fable 5 and Mythos 5 restored
24 July	Claude Opus 5 released
10 August	Sonnet 5 introductory price made permanent
1 September	Claude Fable 5.1 and Mythos 5.1 released
22 September	Claude Opus 5.5 released
28 September	Claude Sonnet 5.5 released

July: access restored, then Opus 5

Anthropic’s statement of 30 June says that on 12 June “the US government applied export controls to our newest models, Claude Fable 5 and Claude Mythos 5.” It adds that Anthropic suspended access for all users because it had no reliable way to verify nationality in real time. The controls were lifted on 30 June, and the release notes record access restored on 1 July.

On 24 July Anthropic released Claude Opus 5 at $5 per million input tokens and $25 per million output tokens, the same price as Opus 4.8. Anthropic says it “comes close to the frontier intelligence of Claude Fable 5 at half the price.” The release notes list a 1M token context window and 128k maximum output.

August: price and platform housekeeping

On 10 August Anthropic made the introductory Sonnet 5 price, $2 / $10 per million tokens, the standard price. A planned rise to $3 / $15 on 1 September “will not occur.”

On the platform side, the computer use tool left beta on 19 August and a separate browser use tool launched. The Files API and Agent Skills also left beta the same day.

September: three model releases in four weeks

Fable 5.1 (1 September). Anthropic says Fable 5.1 and Mythos 5.1 are the same model with different safeguards. Fable 5.1 is generally available; Mythos 5.1 is limited to trusted access programs. Price stays at $10 / $50 per million tokens, but cache reads drop to $0.25. Anthropic says Fable 5.1 “will cost an estimated 25% less than Fable 5 for typical workloads.” Both models carry a text watermark and require 30-day data retention unless Anthropic authorizes otherwise.

Opus 5.5 (22 September). Priced at $4 / $20 per million tokens, with cache reads at $0.20. Anthropic says it costs 40% less than Opus 5 on typical workloads and performs at the level of Fable 5.1 on most work. It also raised five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans.

Sonnet 5.5 (28 September). Same price as Sonnet 5 ($2 / $10, cache reads $0.20). Anthropic says it generates output 30%+ faster and costs up to 30% less per task, because it needs fewer tokens. Anthropic also says “Opus 5.5 remains clearly stronger at complex, open-ended work requiring sustained judgment.”

Anthropic’s own table, as reported on the launch pages:

Benchmark (Anthropic’s figures)	Sonnet 5.5	Opus 5.5	Fable 5.1
Terminal-Bench 4.0	70.6%	66.4%	55.8%
CursorBench 4.0	55.5%	57.8%	51.8%
GDPval-AA v2.1 (Elo)	1844	1846	1735

The Opus 5.5 post adds a caution: “benchmark margins have become a less reliable guide to real-world differences.” Terminal-Bench 4.0 for Opus 5.5 is reported at xhigh effort, and the Fable 5.1 column comes from the Opus 5.5 post, while the Sonnet 5.5 column comes from its own post.

Haiku. The models overview still lists Claude Haiku 4.5 ($1 / $5, 200K context). Anthropic says Claude Haiku 5.5 “will join the Claude 5.5 family in the coming weeks.” It was not released as of 5 October.

API and agent tooling

• Opus 5.5 always uses adaptive thinking; sending thinking: {"type": "disabled"} returns a 400 error. Sonnet 5.5 needs between_tools instead of "disabled".
• tool_choice types any and tool return a 400 error on Fable 5.1, Opus 5.5 and Sonnet 5.5.
• Sonnet 5.5 thinking blocks work only in the account that produced them.
• Beta features added: on-demand conversation compaction (14 September) and tool definitions inside mid-conversation system messages (22 September).
• Claude Managed Agents gained an auto permission policy (10 September) and session budgets (7 August).
• Claude Sonnet 4.5 is deprecated, with retirement scheduled for 30 November 2026.

Claude Code

Claude Code made each model a default on the day it was released: Opus 5 in 2.1.219 (24 July), Fable 5.1 in 2.1.257 (1 September), Opus 5.5 in 2.1.280 (22 September) and Sonnet 5.5 in 2.1.284 (28 September). Other changes in the changelog:

• 2.1.233: the todo and task-tracking tools are no longer available on Opus 4.8, Sonnet 5, Fable 5, Mythos 5 and newer models. Setting CLAUDE_CODE_ENABLE_TODO_TOOLS=1 brings them back.
• 2.1.283: managed settings deniedModels and availableModelsMatch let an organization block specific models, and /doctor prompt-audit audits CLAUDE.md files, skills, agents and commands for patterns written for older models.
• 2.1.287 (1 October): “Claude Mods,” plugins that can modify deeper behavior.

Policy and availability

Opus 5.5 and Sonnet 5.5 ship with cybersecurity safeguards. On Sonnet 5.5, Anthropic says “higher-risk cybersecurity tasks will visibly fall back to Sonnet 5.” On Opus 5.5 most cybersecurity tasks are re-routed to Opus 4.8. Anthropic describes routine software development as unaffected.

On 24 September Anthropic resumed billing for refusals that arrive before any output in the bio, frontier_llm and reasoning_extraction categories.

What this means for a solo developer (my opinion)

Everything below is my reading of the list prices, not a test result.

For a hypothetical task with 100k input and 10k output tokens, ignoring caching and per-task token differences, list prices work out to about $0.15 on Haiku 4.5, $0.30 on Sonnet 5.5, $0.60 on Opus 5.5 and $1.50 on Fable 5.1. That is a tenfold spread, and the models overview itself says to start with Opus 5.5 for most workloads and move to Fable 5.1 only when evals at higher effort still fall short.

My default would be Sonnet 5.5 for everyday coding, with Opus 5.5 reserved for large, ambiguous jobs. Anthropic’s own numbers put Sonnet within a few points of Opus on CursorBench, but they also say Opus is clearly stronger on open-ended work, so I would run my own tasks before trusting a benchmark. Fable 5.1 looks hard to justify at 2.5 times Opus 5.5’s price unless Opus fails on a specific job.

Two practical points. Cache reads are the main cost in agentic work, so the $0.20 rate matters more than the headline price. And the API changes (no disabled thinking, no forced tool_choice) are breaking, so pin model IDs and test before switching a product’s default.