Anthropic shipped Claude Opus 5.5 on 22 Sep 2026. If you already bill client coding against Opus 5, Sonnet 5, or a Claude Pro desk, the launch post’s “40% less on typical workloads” line is the question — not another benchmark dump. Is that claim real for your jobs, or only for cache-heavy agentic runs? Is Fast mode’s $8 / $40 per 1M tokens worth a deadline sprint? Or is waiting for Sonnet 5.5 the smarter wallet move?
This is an upgrade / wait / stay sheet. Not a rewrite of our full Claude vs OpenAI API cost ladder. Cursor project habits stay on the Cursor Projects post. Here the only decision is whether claude-opus-5-5 earns the next client coding run — 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 28 Sep 2026 (SAST) — re-check before you quote a client. Not legal advice; not a guarantee of clients, savings, or income.
What shipped 22 Sep 2026 — Opus 5.5 in plain English
On 22 Sep 2026, Anthropic published Introducing Claude Opus 5.5. It is the first model in the Claude 5.5 family. The Claude API model ID is claude-opus-5-5. Same ID shape on Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS — check each console’s exact string before you swap a production default.
Anthropic positions Opus 5.5 for long-running agentic coding and knowledge work, with a 1M context window and adaptive thinking that is always on. Sonnet 5.5 and Haiku 5.5 are described as coming in the coming weeks, with “many of the same improvements.” Treat that timeline as a vendor claim, not a calendar promise you put in a client SOW.
Biology and cybersecurity safeguards are similar to Anthropic’s other high-capability models. For normal freelance coding — bug fixes, refactors, migrations that stay inside ordinary software work — those gates usually stay out of the way. Cyber-heavy tasks may re-route to Opus 4.8; that is a product note, not a reason to invent win-rates from Anthropic’s competitor tables.
List price vs “40% typical workload” — two different claims
Two numbers get mashed together in secondary posts. Keep them separate.
List price. Opus 5.5 is $4 input / $20 output per 1M tokens on the Claude Platform pricing page. That is about a 20% cut vs Opus 5’s $5 / $25. Cache writes: $5 (5-minute) / $8 (1-hour). Cache reads: $0.20 per 1M — vs Opus 5’s $0.50.
“40% less on typical workloads.” That is Anthropic’s claim on the announcement: list cut plus fewer tokens per task, with cache reads called out as most of the cost on agentic and coding work. It is not a Mr1Tech guarantee that your next invoice drops 40%. One-shot edits with little reuse will feel closer to the ~20% list cut. Cache-heavy loops are where the vendor story is aimed.
Do not invent personal bill savings or ZAR conversions. Quote the USD rows, then meter one real job.
When cache-heavy work makes the upgrade real
The upgrade earns its keep when the same large context keeps coming back: multi-file coding agents, overnight repo runs, long system prompts, and tool loops that re-read the same prefix. Anthropic explicitly frames cache reads as the majority of agentic/coding cost — and Opus 5.5’s $0.20 cache-read row is 60% lower than Opus 5’s $0.50 on the announcement table.
If your last Opus 5 week was mostly re-sending the same codebase scaffold with small deltas, re-running one non-secret job on Opus 5.5 at default effort is the clean A/B. Log token bill and rework minutes. That row beats any launch chart.
When list-price alone is enough — and when Sonnet 5 still wins
For one-off polish, short briefs, and jobs that never warm a cache, the honest gain is closer to the list cut — still useful, not magical. Sonnet 5 remains $2 / $10 per 1M on the same pricing page. If the deliverable already ships on Sonnet 5 inside budget and review time, “upgrade to Opus 5.5 because it launched” is noise.
Waiting for Sonnet 5.5 is rational when your day job is Sonnet-shaped volume and you do not need Opus-tier agentic depth this week. Anthropic says Sonnet 5.5 follows in the coming weeks — still a vendor timeline. Do not pause a paying client on a rumour of a cheaper mid-tier.
Fast mode — when 2.5× speed is worth $8 / $40
Fast mode for Opus 5.5 is listed at $8 input / $40 output per 1M tokens — up to 2.5× speed in Anthropic’s framing. Pricing docs are clear: Fast mode is on the Claude API (first-party) path (and Claude Code / Claude Platform language on the announcement). It is not available on partner-operated Bedrock/GCP paths the same way, and it is not compatible with the Batch API.
Use Fast when latency is burning the fee — a same-day client deadline, a live pairing session, a merge that cannot wait for a standard turn. Skip Fast as a default for overnight batch work; Batch’s 50% discount on input/output is the other lever, and those two do not stack with Fast.
Subscription desk vs API desk
On Pro, Max, Team, and seat-based Enterprise plans, Anthropic says it is increasing five-hour usage limits and providing a rate-limit reset you can bank and use when you choose. Do not invent message caps — quote what your plan screen shows after the bump.
Stay on the subscription desk when Claude is your editor and review partner, not a client-owned API product. Move to the API desk when the SOW needs a client key, a hard USD spend cap, or a harness outside the chat UI. Card and billing-country quirks at Claude Console signup are real for international freelancers — verify at signup; do not invent decline rates. Long agent sessions still need a stable uplink; for mobile-data packaging see fibre vs mobile data for AI work.
Chat-model crown comparisons stay on ChatGPT vs Claude vs Gemini — complementary, not this sheet.
Breaking-change caution before you “just swap the model string”
The Opus 5.5 overview lists real migration edges. Adaptive thinking is always on and cannot be disabled — steer depth with effort, default medium. 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.
Preserved thinking (anti-distillation) applies for API accounts created on or after 31 Aug 2026. If your integration edited prior thinking blocks, read Anthropic’s preserved-thinking docs before you flip the ID in production. Biology/cyber safeguards usually stay irrelevant for ordinary freelance coding; they matter when the brief itself is bio research or cyber offense/defense.
Swapping the string without reading the overview is how you spend a Friday diagnosing silent stream changes between tool calls.
Buy/skip matrix (sheet)
| Job | Switch Opus 5 → 5.5 now | Stay Sonnet 5 | Wait Sonnet 5.5 | Fast mode only | Stay chat Pro (no API) |
|---|---|---|---|---|---|
| Cache-heavy multi-file / overnight agent | Yes — meter one job | Weak | Only if quality already enough | If latency burns the fee | If API isn’t the deliverable |
| One-off polish / short brief | Maybe — list cut only | Often yes | Fine if mid-tier is the plan | Rarely | Yes if chat is the desk |
| Same-day deadline, first-party Claude API | Yes at default | If already shipping | — | Yes — $8/$40 path | Weak for hard latency SLAs |
| Bedrock / GCP partner path only | Standard path — confirm Fast availability | If cheaper enough | Maybe | Usually no on partner paths | — |
| Batch overnight, no urgency | Yes — Batch 50% off | Often enough | Maybe | No — Batch ≠ Fast | — |
| Client SOW locks Sonnet / older ID | Only with change order | Yes | After SOW update | — | If subscription is contracted |
Get the 1-page Opus 5 → 5.5 upgrade 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 coding job, then one sheet row
Pick one non-secret client coding job you already ran on Opus 5 or Sonnet 5. Same brief, same acceptance criteria. Re-run it on Opus 5.5 at default effort — not Fast, unless latency is the billable problem. Compare token bill and rework time. Fill one sheet row: switch now / stay Sonnet 5 / wait Sonnet 5.5 / Fast for deadline sprints only / stay on chat Pro if API isn’t the job.
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 Opus 5.5 announcement, pricing, Opus 5.5 overview — re-check model IDs and Fast / cache rows at paste time.