The Claude Opus price dropped for the first time in the model line's history: Claude Opus 5.5 bills at $4 per million input tokens and $20 per million output tokens, against $5 and $25 for Opus 5 and Opus 4.8 (Anthropic; pricing docs). Verdict: Opus 5.5 wins Opus-tier API work as of September 2026 — it undercuts the model it replaces and beats the $10/$50 Fable 5.1 on Anthropic's published benchmarks, while Sonnet 5 at $2/$10 (pricing docs) still wins high-volume classification where Opus capability is wasted.
TL;DR
- Opus 5.5 (released 22 September 2026): $4 in, $20 out per million, 20% below Opus 5 (Anthropic).
- Cache reads fell 60% to $0.20 per million — the lever that moves real agent bills (Anthropic).
- Opus 4.8 and Opus 5 remain at $5 in, $25 out — the rate every Opus has carried since Opus 4.5 (Anthropic).
- A typical workload runs 40% cheaper end-to-end: fewer tokens per task, cheaper tokens (Anthropic).
- Fable 5.1 at $10/$50 loses its case for cost-conscious teams; Sonnet 5 stays the volume option (pricing docs).
- Last verified: 26 September 2026.
What is the Claude Opus price in 2026?
There are now two Opus rate cards. Opus 5.5 sits at $4 per million input tokens and $20 per million output tokens, with cache writes at $5 per million and cache reads at $0.20 per million (Anthropic). Opus 5 and Opus 4.8 stay on the older card of $5 in, $25 out, with a five-minute cache write at $6.25, a one-hour cache write at $10, and cache hits at $0.50 (Claude pricing docs).
| Model | Input /MTok | Output /MTok | Cache write (5m) | Cache read |
|---|---|---|---|---|
| Claude Opus 5.5 | $4 | $20 | $5 | $0.20 |
| Claude Opus 5 | $5 | $25 | $6.25 | $0.50 |
| Claude Opus 4.8 | $5 | $25 | $6.25 | $0.50 |
| Claude Fable 5.1 | $10 | $50 | $12.50 | $0.25 |
| Claude Sonnet 5 | $2 | $10 | — | — |
| Claude Sonnet 4.6 | $3 | $15 | — | — |
| Claude Haiku 4.5 | $1 | $5 | — | — |
Rates above are from the Claude platform pricing documentation and the Opus 5.5 announcement. One quirk is worth noting: Fable 5.1 and Mythos 5.1 apply a 0.025x cache-hit multiplier rather than the standard 0.1x used across the rest of the range, which is why their cache reads land at $0.25 (pricing docs).
Sticker shock often dates from retired models: Claude Opus 4 and 4.1 listed at $15 in and $75 out per million, nearly four times the current Opus 5.5 input rate (pricing docs).
Why did Anthropic cut the Opus price at all?
Anthropic ties the cut to serving cost: Opus 5.5 needs less compute to run than Opus 5, and the price reflects that (Anthropic). The company also singles out cache reads as the bulk of agentic and coding spend, which is why the 60% reduction there, from $0.50 to $0.20 per million, tends to move a real bill more than the headline input rate does (Anthropic; corroborated by The New Stack).
The other half is behavioural: at default settings a typical workload costs 40% less on Opus 5.5 than Opus 5, because the model spends fewer tokens per task, and output generates more than 30% faster (Anthropic).
Which Opus wins on capability per dollar?
On Anthropic's published numbers, Opus 5.5 reaches 66.4% on Terminal-Bench 4.0 agentic coding against 52.3% for Opus 5 and 55.8% for Fable 5.1, and posts a GDPval-AA v2.1 knowledge-work Elo of 1846 versus 1735 for Fable 5.1 and 1708 for Opus 5 (Anthropic). In a port of HAProxy from C to Rust, Anthropic reports Opus 5.5 finishing in 9.5 hours against 12 hours for Fable 5.1, at 51% lower cost (Anthropic).
Benchmarks are vendor-reported, so treat them as directional. The practical reading is simpler: a model 20% cheaper per token, cheaper on cache reads, and not behind its predecessor removes most reasons to stay on Opus 5. For the cheaper end of the field, see DeepSeek V4.1-Flash vs Opus 5 and GPT-5.6 and Claude Code alternatives.
What is the Claude Opus 4.8 price?
Claude Opus 4.8 costs $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.7, with fast mode at $10 and $50 per million (Anthropic); third-party listings show the same $5/$25 and a $0.50 cache read (OpenRouter). Opus 5 launched at the identical rate (Anthropic). Opus 4.8 keeps its 1M-token context, 128K max output and January 2026 knowledge cutoff, but on price alone Opus 5.5 supersedes it. Free-tier contrast: GLM-5.3-Flash vs Opus 4.8.
How do fast mode, Batch and data residency change the bill?
Three modifiers sit on top of the base rate card:
- Fast mode. Opus 5.5 fast mode runs up to 2.5x speed at $8 input and $40 output per million (Anthropic). On Opus 5 and Opus 4.8 fast mode is $10 and $50 per million, first-party Claude API only, and does not stack with Batch (pricing docs).
- Batch API. A 50% discount applies, putting Opus 5 and Opus 4.8 at $2.50 and $12.50 per million (pricing docs).
- Data residency. US-only inference carries a 1.1x multiplier, so the $5/$25 card becomes $5.50/$27.50 (Anthropic list-price PDF).
One pricing decision that quietly stuck: Sonnet 5's introductory $2/$10 rate is now the standard rate, and the increase to $3/$15 previously scheduled for 1 September 2026 will not happen (pricing docs).
Does the cut reach Pro and Max subscribers?
Indirectly. Claude Pro is $20 a month, Max 5x is $100 and Max 20x is $200 per aggregator eesel; alongside Opus 5.5, Anthropic raised five-hour usage limits on paid plans and added a storable rate-limit reset (Anthropic; The New Stack). Once a seat spills into usage credits, the $4/$20 API card applies. Claude Code pricing across Pro, Max and API maps where each tier wins, and which LLM is best for coding covers cheaper capability tradeoffs.
Why this narrow question? We priced 656 keywords in AI and developer tooling on DataForSEO volume and difficulty data (n=656, measured 21 September 2026); only 72 (11.0%) cleared our winnable bar of 150-6,000 monthly searches, difficulty 20 or below, and genuine technical intent. "Claude Opus price" is one of them.
Limitations worth stating
Opus 5.5 has a June 2026 knowledge cutoff, runs adaptive thinking that cannot be switched off, and carries EU AI Act watermarking (Anthropic). The "40% less" figure assumes Anthropic's default settings and workload mix; replay a week of your own traffic against both model IDs before migrating. AWS, Google Cloud and Azure pricing can diverge from the first-party card (pricing docs).
FAQ
Q: What is the Claude Opus price per million tokens in 2026?
A: Opus 5.5 is $4 input / $20 output per million; Opus 5 and Opus 4.8 remain at $5 / $25, per Anthropic and its pricing docs.
Q: What is the Claude Opus 4.8 price?
A: $5 input / $25 output per million, fast mode at $10 / $50, unchanged from Opus 4.7 per Anthropic's announcement.
Q: Is Opus 5.5 cheaper than Fable 5.1?
A: Yes: $4/$20 against Fable 5.1's $10/$50, and Anthropic's benchmarks put Opus 5.5 ahead on agentic coding and knowledge work.
Q: Why did cache pricing change so much?
A: Anthropic says cache reads dominate agentic and coding spend, so they were cut hardest: $0.20 per million from $0.50.
Q: Should I switch from Sonnet 5 to Opus 5.5?
A: Not for high-volume, low-judgement work: Sonnet 5 at $2/$10 (pricing docs) stays cheaper for classification, extraction and summarisation.
Q: Does the cut apply to Claude Pro and Max subscriptions?
A: Subscription prices held, but five-hour limits rose and overflow now bills at $4/$20 instead of $5/$25 (Anthropic).
Last verified: 26 September 2026. Corrections and rate changes will be noted here.
Top comments (0)