The flagship got it twelve days earlier
Anthropic gave Claude Opus 4.6 a beta one million token context window on February 5, 2026. Twelve days later, Sonnet 4.6 shipped with the same feature: Anthropic's announcement states the model "also features a 1M token context window in beta." [1]
This is not the first time Sonnet has had a million tokens of context. Claude Sonnet 4 picked up the same beta feature back on August 12, 2025, five months before Opus 4.6 got it. Neither Anthropic's Sonnet 4.6 post nor Simon Willison's coverage of it draws that connection explicitly, but the September 2025 Sonnet 4.5 release did not carry the feature forward in its own announcement, and the February 2026 post treats the million-token window as new to Sonnet 4.6 rather than as a restoration. [1][2]
Atlas interpretation: The sequence, Sonnet first in 2025, then Opus, then Sonnet again in 2026, matters less than the gap between the two 2026 releases. Twelve days is a short enough interval that the capability difference between the flagship and the mid-tier model, on this one dimension, briefly amounted to less than two weeks of Anthropic's release calendar. [1]
Capability closed the gap. Price did not.
Anthropic's post says pricing for Sonnet 4.6 "remains the same as Sonnet 4.5, starting at $3/$15 per million tokens." Opus models are priced at $5 per million input tokens and $25 per million output tokens, per Willison's independent summary of the same-day release. [1][2]
Willison's post notes that both the Sonnet and Opus lines cap at 200,000 tokens by default and can be stretched to the million-token beta window only "at a higher cost." Anthropic's own announcement does not state that surcharge for Sonnet 4.6, and gives Opus 4.6 tiered per-token pricing for prompts longer than 200,000 tokens rather than a single flat rate. [2]
Atlas interpretation: Sonnet catching up on context length without a headline price change is the part of the release that matters more than the number itself. A gap that used to separate what the flagship model could hold in memory from what the cheap model could hold is gone, at least in beta, while the price difference between the two tiers has not moved. That is what closes the distance between what a team can justify building on the inexpensive model versus the expensive one. [1][2]
Preferred to the model that used to cost more
Anthropic reports that developers with early access preferred Sonnet 4.6 to Sonnet 4.5 "by a wide margin," and that in Claude Code testing users preferred it over Sonnet 4.5 roughly 70 percent of the time. The same post says users preferred Sonnet 4.6 to Claude Opus 4.5, the November 2025 flagship, 59 percent of the time, citing better instruction-following and fewer hallucinations. [1]
Willison's independent take frames the same comparison from the other direction: Sonnet 4.6 offers "similar performance to November's Opus 4.5" while keeping Sonnet's lower price. He also compares image-generation quirks between Sonnet 4.6 and Opus 4.6 directly, judging Opus 4.6's pelican illustrations technically better while noting Sonnet 4.6's own touches. [2]
Atlas interpretation: A preference rate against the previous flagship is Anthropic's own comparison, run on its own testers, and it is not the same claim as beating Opus 4.5 on a fixed benchmark. But it is a specific, falsifiable number rather than marketing language, and an independent writer arrived at a compatible read of the same release without citing that figure. Both point the same direction: the mid-tier model closed distance on the one above it, in the reasoning users reward as well as in raw context length. [1][2]
Sources
- Introducing Claude Sonnet 4.6
Anthropic · Feb 17, 2026
- Claude Sonnet 4.6
Simon Willison · Feb 17, 2026