Anthropic shipped Claude Sonnet 5.5 on 28 Sep 2026 — one week after Opus 5.5. If you already bill everyday coding, bug fixes, and polished docs against Sonnet 5, the launch post’s “up to 30% less per task” and “30%+ faster” lines are the question — not another benchmark dump. Same $2 / $10 list price as Sonnet 5. Different token bill if Anthropic’s fewer-tokens claim holds on your jobs. Or is Opus still the right desk for open-ended judgment work?
This is a switch / skip Opus / wait / stay sheet. Not a rewrite of our Claude Opus 5.5 upgrade post — that sheet covers Opus Fast mode, Opus cache story, and when Opus earns the seat. Full Claude vs OpenAI ladder stays on API cost for freelancers. Here the only decision is whether claude-sonnet-5-5 replaces Sonnet 5 as your mid-tier default — now, later, or not at all.
Affiliate disclosure: some tool or platform links may later be affiliate or referral. Prices below are from Anthropic’s live announcement and Claude Platform pricing docs as of 29 Sep 2026 (SAST) — re-check before you quote a client. Not legal advice; not a guarantee of clients, savings, or income.
What shipped 28 Sep 2026 — Sonnet 5.5 in plain English
On 28 Sep 2026, Anthropic published Introducing Claude Sonnet 5.5. It is the second model in the Claude 5.5 family, after Opus 5.5. The Claude API model ID is claude-sonnet-5-5. Same ID shape on Bedrock (anthropic.claude-sonnet-5-5), Google Cloud, Microsoft Foundry, and Claude Platform on AWS — confirm each console’s exact string before you swap a production default.
Anthropic positions Sonnet 5.5 as a faster, lower-cost complement to Opus 5.5. Where Opus handles complex work that needs careful judgment, Sonnet 5.5 is framed for well-scoped everyday tasks, bug fixes, polished documents, slides, and spreadsheets — plus a sharper design eye. Context window is 1M; max output 128K on the Messages API (Batch can go to 300K with the documented beta header). Adaptive thinking is on; Claude API default effort is high (apps default Medium on the announcement). Haiku 5.5 is described as joining “in the coming weeks” — treat that as a vendor timeline, not a client SOW date.
Biology safeguards match Sonnet 5. Cyber safeguards are elevated (fallback to Sonnet 5 on higher-risk cyber). For normal freelance coding — fixes, refactors, decks that stay inside ordinary software and knowledge work — those gates usually stay out of the way. Zero data retention is available like Opus 5.5 and Sonnet 5. For Opus-shaped architecture weeks, keep using the Opus 5.5 freelancer upgrade sheet — do not re-argue Fast mode here.
Same list price, different bill
Two numbers get mashed together in secondary posts. Keep them separate.
List price. Sonnet 5.5 is $2 input / $10 output per 1M tokens on the Claude Platform pricing page — unchanged vs Sonnet 5. Cache reads: $0.20. Cache writes: $2.50 (5-minute) / $4 (1-hour). Batch API is 50% off → $1 / $5. Opus 5.5 contrast on the same page: $4 / $20.
“Up to 30% less per task” / “30%+ faster.” Those are Anthropic’s claims on the announcement: fewer tokens per task in their testing, plus faster output generation. They are not a Mr1Tech guarantee that your next invoice drops 30%. Quote the USD rows, then meter one real job. Do not invent personal bill savings, ZAR conversions, or Claude Pro message quotas. Fast mode at $8 / $40 is documented for Opus paths only — there is no Sonnet Fast tier to invent.
Switch from Sonnet 5 now — when the upgrade is boring and correct
The upgrade is boring and correct when your week is already Sonnet-shaped: everyday coding, iterative bug fix, docs/slides/spreadsheets polish, Cursor-style multi-file work that stays well-scoped. Anthropic’s coding and knowledge-work sections on the announcement describe early-tester themes — fewer tool steps, less web-search thrash, clearer writing, sharper design polish. Useful direction. Not a win-rate you paste into a client proposal.
Meter one non-secret job at default effort before flipping production defaults. Same brief, same acceptance criteria you already shipped on Sonnet 5. Log token bill, rework minutes, and wall time. If the meter is quieter and the deliverable still passes review, swapping the model ID at the same $2 / $10 list is the low-drama move.
Skip inventing “average freelancer Sonnet bills.” One logged row beats a launch chart. If quality already ships inside budget on Sonnet 5 and you refuse a one-job A/B this week, staying is also a real choice — see Wait / stay below.
Skip Opus (for this job) — when Sonnet 5.5 is enough
Opus remains for complex open-ended judgment and architecture. Anthropic is explicit: even when Sonnet scores near Opus on some benches, Opus stays clearly stronger on that class of work. Freelancer rule: if the SOW is well-scoped implementation after you already set the design, Sonnet 5.5 is the wallet default at half Opus list ($2 / $10 vs $4 / $20).
A clean split many desks already use: Opus plans the architecture; Sonnet implements. That pattern does not need a Fast-mode argument or a rewrite of Opus cache economics — those live on the Opus upgrade sheet. This post only answers whether this job needs Opus at all.
Skip Opus for this job when the brief is implementation after a locked design, same-day iterative fix, overnight Batch at $1 / $5, or deck/doc polish that does not need sustained open-ended judgment. Keep Opus when the week is architecture, ambiguous SOW, or long judgment loops where half-price mid-tier would cost you more in rework than it saves on tokens.
Wait / stay — when not to touch the default yet
Wait when the client SOW locks an older model ID — change order first, not a silent string swap. Wait when you have not metered one job yet; a launch post alone is not a production flip mid-client week. Wait when your partner Bedrock/GCP/Foundry string is not confirmed in your console.
The Sonnet 5.5 overview lists real migration edges vs Sonnet 5. If you ran Sonnet with thinking off, switch to between_tools before moving to 5.5. Forced tool use returns an error. Thinking blocks are tied to the model and conversation. On the Claude API and Google Cloud, the earlier computer_20251124 computer-use tool is not accepted — use the current toolset docs. Text between tool calls can land in thinking blocks and go quiet in streams until you set a display value or use between_tools.
Stay on Sonnet 5 only if quality already ships inside budget and you refuse a one-job A/B this week. Same list price means “wait for a discount” is not the reason — migration caution or SOW lock is. Chat-model crown comparisons stay on ChatGPT vs Claude vs Gemini — complementary, not this sheet. Cursor folder/rules habits stay on Cursor Projects; this section is only which model ID you point at the job.
Buy/skip matrix (sheet)
| Job | Switch Sonnet 5 → 5.5 now | Skip Opus — use Sonnet 5.5 | Keep Opus for judgment | Stay Sonnet 5 | Wait (SOW / migration) |
|---|---|---|---|---|---|
| Everyday coding / well-scoped multi-file | Yes — meter one job | Yes if design already set | Only if open-ended | If you refuse A/B | If partner ID unconfirmed |
| Docs / deck / spreadsheet polish | Yes — meter one job | Usually yes | Rarely | If already shipping | — |
| Overnight batch (Batch 50%) | Yes — $1 / $5 path | Yes | Weak for Batch volume | Same Batch rates | — |
| Same-day iterative fix | Yes at default | Yes | If judgment is the bottleneck | If already inside budget | — |
| Architecture + implement split | Sonnet implements | Implement on Sonnet | Opus plans | — | If SOW locks IDs |
| Client SOW locks older model ID | Only with change order | — | If contracted | Yes until SOW update | Yes — change order first |
Get the 1-page Sonnet 5 → 5.5 / skip Opus / wait / stay chooser
Optional printable of this free guide. Soft link until checkout. Not a “paid summary.” Not a guarantee of clients, savings, or income.
This week’s action — one non-secret Sonnet-shaped job
Pick one non-secret Sonnet-shaped job you already ran on Sonnet 5. Same brief, same acceptance criteria. Re-run it on claude-sonnet-5-5 at default effort. Compare token bill, rework minutes, and wall time. Fill one sheet row: switch Sonnet 5 → 5.5 now / skip Opus use Sonnet 5.5 / keep Opus for judgment work / stay Sonnet 5 / wait for SOW or migration edges.
If the meter surprises you, the surprise is the product — not a reason to hide usage from the next quote. Do not change a production default mid-client week on a launch post alone.
Sources: Claude Sonnet 5.5 announcement, pricing, Sonnet 5.5 overview — re-check model IDs, effort defaults, and cache rows at paste time.