Claude Sonnet 5.5 for Marketers: A First-Week Take on What "Near-Opus at Sonnet Pricing" Actually Buys You

The Friday before Sonnet 5.5 shipped, I ran a 12-touch B2B nurture sequence through Sonnet 5 at default effort to set a baseline. Re-running the same sequence on Sonnet 5.5 the day it shipped, at default effort, landed the first draft in 38% less wall time and used 24% fewer output tokens — close to Anthropic's "up to 30%" claim, on real marketing work. Within a day of use, I'd already moved four of my daily pipelines off Sonnet 5 onto Sonnet 5.5.

Anthropic shipped Claude Sonnet 5.5 on Monday, September 28, 2026 — the second model in the new Claude 5.5 family, six days after Opus 5.5. Anthropic's framing: Sonnet 5.5 "approaches Opus-class performance" at Sonnet pricing, with the model typically needing fewer tokens per task. After a first day of real marketing work, the framing is honest but undersells the upgrade in the wrong direction. The headline isn't "30% cheaper." It's that the Sonnet tier can now carry work I previously reserved for Opus 5.5 — and the routing math for marketers shifts more than the sticker price suggests.

What Sonnet 5.5 is, briefly

• API model id: claude-sonnet-5-5. Live in Claude API, AWS Bedrock, Google Cloud, Claude Code, the apps.
• Context: 1M input / 128K output. Same window as Sonnet 5, Opus 5, Opus 5.5, Fable 5.1. No long-context surcharge.
• Pricing: $2 input / $10 output per million tokens — same list as Sonnet 5 and as GPT-6 Sol, OpenAI's mid-tier that dropped the same day as Opus 5.5. Cache reads $0.20/MTok, writes $2.50/MTok.
• Speed & cost: Output is 30%+ faster than Sonnet 5 (Anthropic's fastest Sonnet to date) and up to 30% cheaper per task because the model needs fewer tokens for the same work.
• Thinking: Adaptive, always on, cannot be disabled. Default effort is medium in Claude Code and the apps, high on the Claude Platform / API — a behavior split worth knowing. Five effort levels: low / medium / high / xhigh / max.
• Safeguards: First Sonnet with Fable-class cybersecurity classifiers and model fallbacks. Some cyber-flavored requests silently re-route to Sonnet 5 — check your day-1 logs.
• Watermarking: Ships with invisible text watermarking (EU AI Act compliance). Doesn't change readability.

Model	Input	Output	Cache read	Default effort	Where it earns its slot
Fable 5.1	$10	$50	$0.25	high	Longest-context synthesis, 1M-token recall
Opus 5.5	$4	$20	$0.20	medium	Agentic orchestration, complex research, long outputs
Sonnet 5.5	$2	$10	$0.20	medium	Mid-complexity production work, daily pipelines
Sonnet 5	$2	$10	$0.20	high	Default-tier high-volume (now displaced for most production work)
Haiku 5	$1	$5	$0.10	high	Cheap tier

Same dollar cost as Sonnet 5; meaningfully better output; same cache math. Sonnet 5.5 is not a price event — it's a quality-per-token event.

The benchmark card, with the footnotes that matter

• GDPval-AA v2.1 (real work artifacts, 44 occupations): Sonnet 5.5 at 1,844 Elo — two points under Opus 5.5's 1,846, about 400 points above Sonnet 5's 1,449. That alone reorders the routing: Sonnet 5.5 handles knowledge-work prompts that previously needed Opus.
• Terminal-Bench 4.0: 70.6% at its best effort — beats Opus 5.5's 66.4%, the highest Sonnet-class score Anthropic has published.
• Humanity's Last Exam (with tools): 64.5% — close to Opus 5.5. First Sonnet to beat Pokémon Red from screenshots.
• At low or medium, beats Sonnet 5's best for ~1/10 the cost per task on multiple benchmarks. The effort dial matters more than I expected.

Two caveats: the cyber classifier may silently route a request to Sonnet 5 — don't assume every "Sonnet 5.5" call hit the new model. And on the absolute hardest reasoning, Fable 5.1 still leads. Anthropic's framing — "Opus 5.5 remains clearly stronger at complex, open-ended work requiring sustained judgment" — is accurate; it just describes a smaller lane than I expected.

Five jobs Sonnet 5.5 already took from me since it shipped

1. The 12-touch B2B nurture sequence

I ran the same sequence on Sonnet 5 (default) and Sonnet 5.5 (default) back-to-back since Sonnet 5.5 shipped. Sonnet 5.5 landed the first draft in 38% less wall time and used 24% fewer output tokens. Subject-line A/B variants read better out of the box; the third email, which previously needed an edit pass to fix a tonal drift, landed clean. For a sequence I run weekly across six clients, that's the kind of workflow where 30% per-task savings compounds.

2. Daily Reddit monitor escalation layer

The Reddit monitoring agent runs Sonnet for daily triage and only escalates to Opus for high-signal items. Sonnet 5.5 at default now skips escalation about 60% of the time — up from 45% on Sonnet 5. Fewer Opus calls, lower bill, faster daily digest.

3. RSS-to-Newsletter pipeline

Newsletter curation with Claude + RSS has been a Sonnet-tier job for years. The metric I track is "paragraphs that ship without an edit." Sonnet 5.5 cleared 22 of 30 since Sonnet 5.5 shipped; Sonnet 5 cleared 14 on a comparable baseline run. The work the pipeline does didn't change — the work the brand-voice analyzer doesn't have to catch changed.

4. First-pass campaign post-mortem assembly

The campaign post-mortem template has eight sections. Sonnet 5.5 at medium produces a draft where I only need to rewrite section 3 (the one that depends on the brief's positioning). Sonnet 5 at the same effort needed three rewrites. Opus 5.5 still wins the final version, but the first-pass saves 20 minutes.

5. Slack /research cited briefs

The /research command needs a model that cites, doesn't hallucinate, and answers in three paragraphs. Sonnet 5.5 at medium does the job at half the cost of Opus 5.5 with cleaner prose than Sonnet 5. The upgrade most visible to my team — they don't notice the bill, they notice the brief reads like a person wrote it.

Three jobs where Opus 5.5 or Sonnet 5 still wins

1. The longest, hardest-context reasoning — Opus 5.5 (or Fable 5.1). Reading a 1M-token pile in one session and asking for synthesis without chunking, Opus 5.5 still wins. The 240-page quarterly competitor dump → strategy brief, the multi-day agent session that needs persistent context — those stay on Opus 5.5 or Fable 5.1. Sonnet 5.5 isn't far behind, but "near-Opus" isn't "Opus," and the gap on the hardest 5% of jobs is the gap that matters.

2. Regulated refusal and "I need it to push back" — Opus 5.5 or Fable 5.1. In regulated verticals — financial advice, medical copy, anything EU AI Act gated — keep Opus 5.5 or Fable 5.1 for the first 90 days. The new Fable-class cyber safeguards on Sonnet 5.5 aren't fully public-tested in customer deployments yet. Watch item, not blocker, but don't migrate regulated pipelines on day one.

3. The cheap tier where Sonnet 5 used to live — wait for Haiku 5.5. Sonnet 5 isn't dead, but harder to justify. Sonnet 5.5 at low beats Sonnet 5 at any setting on multiple benchmarks, for the same dollar. One place I'd still pick Sonnet 5: legacy pipelines you can't risk re-validating this week. My re-validation of three pipelines found one edge case (multi-column long-context PDF extraction) where Sonnet 5.5 at medium produced different output formatting. Test before swapping.

The routing principle, after a day on real marketing work

Sonnet 5.5 is the new default for mid-complexity production work. Opus 5.5 stays for the hardest-context reasoning and agentic orchestration. Fable 5.1 stays for the absolute ceiling. Sonnet 5 gets displaced to legacy.

In practice:

• Sonnet 5.5 at medium — the 12-touch nurture, daily Reddit monitor, RSS-to-newsletter pipeline, first-pass campaign post-mortem, Slack /research briefs, daily competitor ad sweep, RSA variations, subject lines.
• Opus 5.5 at medium — the 240-page competitor dump → strategy brief, multi-agent orchestration, the multi-agent competitive intel pipeline, long outputs where the first paragraph has to land.
• Fable 5.1 at high — the 1M-token synthesis without chunking.
• Sonnet 5 (legacy) — only where re-validating Sonnet 5.5 is a risk you can't take this week.
• Haiku 5 (later Haiku 5.5) — the cheap tier when any of the above is overkill.

The bit Anthropic got right that most coverage missed: Sonnet 5.5 isn't a cheaper Opus. It's a Sonnet-tier model that can carry real marketing work for the same dollar, and the displacement happens at the Sonnet-5-vs-Sonnet-5.5 line, not at the Opus tier. The Reddit monitor that used to escalate to Opus on every high-signal day now mostly stays on Sonnet. The campaign post-mortem that needed Opus for the first pass now starts on Sonnet 5.5.

Watch for the silent cyber-routing behavior in your day-1 logs, and run your three highest-volume Sonnet-5 pipelines through Sonnet 5.5 this week — the per-task savings alone are worth the migration.