What Happened
Anthropic has removed Claude Sonnet 4.5 from both the Claude web app and desktop app as of March 6, 2026. Users who relied on the older model no longer have the option to select it in the interface. The only Sonnet-tier model now available through Claude's consumer products is Sonnet 4.6.
The change was made without a formal announcement. Users discovered it when the model simply disappeared from the model selector. API access status for Sonnet 4.5 remains unclear, but the consumer-facing products have been cleaned up to show only the current generation.
This follows Anthropic's typical deprecation pattern. When the Claude 4 family launched, older 3.5 models were phased out on a similar timeline. Sonnet 4.5 had been available alongside 4.6 for several weeks as a transition period, and that window has now closed.
Why It Matters
For most users running standard productivity tasks - summarization, analysis, coding assistance - this change is invisible. Sonnet 4.6 benchmarks higher across most measurable metrics, and for structured work it performs well.
The pain point is specific but real: creative writing users are reporting noticeable regressions with Sonnet 4.6. The complaints center on three issues. First, the model reasons less deeply about narrative and character choices. Second, outputs contain more generic phrasings that feel templated. Third, prompt adherence for complex creative instructions is weaker than what 4.5 delivered.
Some users have found partial workarounds. Adding "think harder" or "ultrathink" to prompts can trigger more deliberate responses from 4.6, but results are inconsistent. This is essentially asking the model to use more of its extended thinking capacity, which helps but doesn't fully replicate the 4.5 behavior.
For teams that built creative workflows around Sonnet 4.5's specific characteristics, this is a forced migration with no opt-out on the consumer side.
Our Take
Model deprecation is inevitable, and Anthropic can't maintain every version forever. But the gap between "technically better benchmarks" and "actually better for my use case" is something the AI industry keeps underestimating.
Sonnet 4.6 is objectively a stronger model on paper. But creative writing quality is subjective and hard to benchmark. When users say the outputs feel more generic, that's a real quality signal that doesn't show up in MMLU scores.
The practical advice: if creative writing quality matters to your workflow and you have API access, check whether Sonnet 4.5 is still available there. If not, invest time in prompt engineering for 4.6 rather than fighting the deprecation. Extended thinking prompts ("think step by step about the narrative choices before writing") tend to produce better creative output than the default behavior.
Long term, this is a reminder not to build rigid workflows around a specific model version. Every model gets deprecated eventually. The users who adapted fastest were the ones who already had prompt templates they could tune, not ones who relied on a particular model's default personality.