Claude API Pricing 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: October 1, 2026 | License: CC BY 4.0
Claude Opus 5.5, Anthropic’s recommended default model since its September 22, 2026 launch, lists at $4 per million input tokens and $20 per million output tokens on the Claude API, per Anthropic’s pricing documentation (verified October 1, 2026). Claude Fable 5.1 lists at $10 / $50, Sonnet 5.5 at $2 / $10 and Haiku 4.5 at $1 / $5.
Quick Answer: How Much Does the Claude API Cost?
Claude API pricing in October 2026 runs from $1 to $10 per million input tokens and $5 to $50 per million output tokens across Anthropic’s four current models (Haiku 4.5 to Fable 5.1). Batch halves every rate. Axis Intelligence Research’s Tokenizer-Adjusted Workload Rate (TAWR) puts a typical support chatbot at $1.25 (Haiku 4.5), $3.26 (Sonnet 5.5), $6.41 (Opus 5.5) and $15.89 (Fable 5.1) per million words-equivalent of traffic.
Key Findings
- Claude Opus 5.5 costs 67% less than Claude Opus 4.1 on a coding-agent workload once tokenizer inflation is priced in, according to Axis Intelligence Research’s TAWR ($4.93 vs $14.74), calculated from Anthropic’s October 1, 2026 rate card.
- Sonnet 5.5’s 33% sticker discount versus Sonnet 4.6 shrinks to about 13% in practice, Axis Intelligence Research estimates, because Claude 4.7-and-later models produce roughly 30% more tokens for the same text, per Anthropic’s pricing documentation.
- Cache reads on Claude Fable 5.1 cost 2.5% of the input rate ($0.25 per million tokens), down from $1 on Fable 5 — a 75% cut announced September 1, 2026, per Anthropic.
- Opus 5.5 cut cache reads 60%, from $0.50 to $0.20 per million tokens, at its September 22, 2026 launch, per Anthropic’s announcement — larger than its 20% cut on input and output.
- Haiku 4.5, Anthropic’s cheapest current model, can be retired as soon as October 15, 2026, per Anthropic’s model-deprecations page, while Anthropic says Haiku 5.5 will ship “in the coming weeks.”
Claude API Pricing per 1M Tokens: The Current Lineup
Four models make up Anthropic’s current lineup. Anthropic’s model overview now tells developers to start with Opus 5.5 for most workloads and reserve Fable 5.1 for demanding reasoning and long-horizon agentic work — a notable shift, because it means the recommended default is no longer the most expensive model.
Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5 pricing (standard, per million tokens)
| Model | API ID | Input | 5-min cache write | 1-hour cache write | Cache read | Output | Context | Verified | Source |
|---|---|---|---|---|---|---|---|---|---|
| Claude Fable 5.1 | claude-fable-5-1 | $10.00 | $12.50 | $20.00 | $0.25 | $50.00 | 1M | 2026-10-01 | Anthropic |
| Claude Opus 5.5 | claude-opus-5-5 | $4.00 | $5.00 | $8.00 | $0.20 | $20.00 | 1M | 2026-10-01 | Anthropic |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | $2.00 | $2.50 | $4.00 | $0.20 | $10.00 | 1M | 2026-10-01 | Anthropic |
| Claude Haiku 4.5 | claude-haiku-4-5-20251001 | $1.00 | $1.25 | $2.00 | $0.10 | $5.00 | 200K | 2026-10-01 | Anthropic |
Context windows and API IDs come from Anthropic’s models overview. Fable 5.1, Opus 5.5 and Sonnet 5.5 all cap output at 128K tokens; Haiku 4.5 at 64K.
Two details in this table matter more than the headline rates. First, Sonnet 5.5 and Opus 5.5 share the same $0.20 cache-read price even though Opus costs twice as much per fresh input token. For agent loops that spend most of their tokens re-reading context, that collapses much of the gap between the two models. Second, there is no long-context surcharge: Anthropic bills Claude 4.6-and-later models at standard rates across the full 1M-token window, so a 900K-token prompt costs the same per token as a 9K-token one.
Other Claude models still on the API
| Model | Input | Cache read | Output | Status (Oct 1, 2026) | Source |
|---|---|---|---|---|---|
| Claude Mythos 5.1 | $10.00 | $0.25 | $50.00 | Active, restricted access | Anthropic |
| Claude Fable 5 / Mythos 5 | $10.00 | $1.00 | $50.00 | Active | Anthropic |
| Claude Opus 5 | $5.00 | $0.50 | $25.00 | Active | Anthropic |
| Claude Opus 4.8 / 4.7 / 4.6 / 4.5 | $5.00 | $0.50 | $25.00 | Active | Anthropic |
| Claude Sonnet 5 | $2.00 | $0.20 | $10.00 | Active | Anthropic |
| Claude Sonnet 4.6 | $3.00 | $0.30 | $15.00 | Active | Anthropic |
| Claude Sonnet 4.5 | $3.00 | $0.30 | $15.00 | Deprecated; retires Nov 30, 2026 | Anthropic |
| Claude Haiku 3.5 | $0.80 | $0.08 | $4.00 | Retired Feb 19, 2026 | Anthropic |
All prices from Anthropic’s pricing page; status from Anthropic’s model deprecations page, both read October 1, 2026. Haiku 3.5 still appears on the pricing table even though the deprecations page lists it as retired — we report both as displayed.
Claude Batch API, Fast Mode and Data Residency: Every Price Modifier
The base rate card is only the starting point. Five modifiers move a Claude bill, and they stack multiplicatively.
| Modifier | Effect on price | Applies to | Source |
|---|---|---|---|
| Batch API | 0.5× input and output | All models; not fast mode; not Managed Agents | Anthropic |
| 5-minute cache write | 1.25× base input | All models | Anthropic |
| 1-hour cache write | 2× base input | All models | Anthropic |
| Cache read | 0.1× base input (0.05× Opus 5.5; 0.025× Fable 5.1 and Mythos 5.1) | All models | Anthropic |
US-only inference (inference_geo: "us") | 1.1× on every token category | Claude 4.6 and later | Anthropic |
| Fast mode | Opus 5.5: $8 / $40; Opus 5 and 4.8: $10 / $50 | Claude API only | Anthropic |
Claude Batch API pricing (50% off)
| Model | Batch input | Batch output | Source |
|---|---|---|---|
| Claude Fable 5.1 | $5.00 | $25.00 | Anthropic |
| Claude Opus 5.5 | $2.00 | $10.00 | Anthropic |
| Claude Sonnet 5.5 | $1.00 | $5.00 | Anthropic |
| Claude Haiku 4.5 | $0.50 | $2.50 | Anthropic |
The Batch API documentation says most batches finish within an hour and every batch has a 24-hour window, with a ceiling of 100,000 requests or 256 MB per batch. For nightly classification, eval runs or document backfills, Batch is the single largest discount available without negotiation.
Is Claude fast mode worth paying double?
Fast mode for Opus 5.5 costs $8 per million input and $40 per million output — exactly 2× the standard rate — for up to 2.5× higher output tokens per second, per Anthropic’s fast mode documentation. The math is clean: you pay a 100% premium for up to a 150% throughput gain. It earns its keep where a human is watching a long agent run; it is wasted on anything asynchronous, and Anthropic does not allow it with Batch.
Opus 5.5’s fast mode is also cheaper than Opus 5’s ($10 / $50), so teams that adopted fast mode on Opus 5 get a 20% cut simply by switching model IDs.
The Tokenizer-Adjusted Workload Rate (TAWR): What Claude Actually Costs Per Unit of Text
Here is the line most Claude pricing pages skip. Anthropic’s pricing documentation states that Claude 4.7-and-later models use a newer tokenizer that produces approximately 30% more tokens for the same text, with the exact increase depending on content. Sonnet 4.6 and earlier use the previous tokenizer.
That means a per-token price comparison across Claude generations is not a like-for-like comparison. Sonnet 5.5 at $2 per million tokens is not 33% cheaper than Sonnet 4.6 at $3 for the same document, because the same document becomes more tokens on Sonnet 5.5.
So Axis Intelligence Research built a number that corrects for it.
Tokenizer-Adjusted Workload Rate (TAWR): the blended cost, in US dollars, of processing the amount of text that Claude’s previous-generation tokenizer would count as 1 million tokens, for a defined workload mix, weighting each Anthropic price component by its share of that workload.
Formula: TAWR = [(s_input × P_input) + (s_read × P_cache_read) + (s_write × P_cache_write_5m) + (s_output × P_output)] × k
where s is each token type’s share of traffic (summing to 1), P is Anthropic’s list price per million tokens, and k is the tokenizer factor: 1.3 for Claude 4.7-and-later models (Anthropic’s “approximately 30%” figure) and 1.0 for Sonnet 4.6, Haiku 4.5 and earlier.
The three workload profiles
| Profile | Fresh input | Cache read | Cache write | Output | Typical shape |
|---|---|---|---|---|---|
| Support chatbot | 40% | 40% | 5% | 15% | Shared system prompt, short replies |
| RAG / document Q&A | 75% | 10% | 5% | 10% | New retrieved chunks every call |
| Coding agent | 10% | 70% | 5% | 15% | Long, repeated context across loop steps |
These shares are Axis Intelligence Research modeling assumptions, identical to those behind the Effective Workload Rate on our OpenAI API pricing analysis, so the two pages can be read side by side. Substitute your own shares; every input is in the dataset.
TAWR by model and workload (standard tier, October 1, 2026)
| Model | Support chatbot | RAG / document Q&A | Coding agent | Tokenizer factor | Source |
|---|---|---|---|---|---|
| Claude Fable 5.1 | $15.89 | $17.10 | $12.09 | 1.3 | Axis Intelligence Research (TAWR) |
| Claude Opus 5.5 | $6.41 | $6.85 | $4.93 | 1.3 | Axis Intelligence Research (TAWR) |
| Claude Sonnet 5.5 | $3.26 | $3.44 | $2.55 | 1.3 | Axis Intelligence Research (TAWR) |
| Claude Haiku 4.5 | $1.25 | $1.32 | $0.98 | 1.0 | Axis Intelligence Research (TAWR) |
| Claude Opus 5 | $8.14 | $8.60 | $6.39 | 1.3 | Axis Intelligence Research (TAWR) |
| Claude Sonnet 4.6 | $3.76 | $3.97 | $2.95 | 1.0 | Axis Intelligence Research (TAWR) |
| Claude Opus 4.1 (retired) | $18.79 | $19.84 | $14.74 | 1.0 | Axis Intelligence Research (TAWR) |
Worked example, Opus 5.5 coding agent: (0.10 × $4) + (0.70 × $0.20) + (0.05 × $5) + (0.15 × $20) = $0.40 + $0.14 + $0.25 + $3.00 = $3.79 per million tokens; × 1.3 = $4.93 per million previous-generation-tokens of text.
Three things fall out of this table.
The Sonnet discount is smaller than it looks. On sticker price, Sonnet 5.5 undercuts Sonnet 4.6 by 33%. On TAWR, the chatbot gap is $3.26 against $3.76 — about 13%. Teams budgeting a Sonnet 4.6 → 5.5 migration from the rate card alone will overestimate their savings by more than half. Anthropic’s own claim, that Sonnet 5.5 costs up to 30% less per task than Sonnet 5, rests on the model needing fewer tokens to finish work, which no price table can show and which you should verify on your own traffic.
Haiku 4.5 is the quiet value outlier. Because it still runs on the older tokenizer, its effective rate is roughly 2.6× cheaper than Sonnet 5.5 on chatbot traffic, against a 2× gap on the sticker. That matters for routing — and it is precisely the model on the shortest retirement clock.
Opus is where the deflation is real. On a coding-agent workload, Opus 5.5’s TAWR of $4.93 is 67% below Opus 4.1’s $14.74 and 23% below Opus 5’s $6.39, even after paying the tokenizer tax. The 60% cache-read cut does most of that work: in an agent loop where 70% of traffic is cache reads, cache pricing is the price.
What does 1 billion tokens of traffic a month cost on Claude?
| Model | TAWR (support chatbot) | Monthly cost, 1B tokens-equivalent | 3-year cost at that volume | Source |
|---|---|---|---|---|
| Claude Fable 5.1 | $15.8925 | $15,892.50 | $572,130 | Axis Intelligence Research estimate |
| Claude Opus 5.5 | $6.409 | $6,409.00 | $230,724 | Axis Intelligence Research estimate |
| Claude Sonnet 5.5 | $3.2565 | $3,256.50 | $117,234 | Axis Intelligence Research estimate |
| Claude Haiku 4.5 | $1.2525 | $1,252.50 | $45,090 | Axis Intelligence Research estimate |
Method: TAWR × 1,000 × months, holding October 1, 2026 list prices constant. Anthropic has cut Opus-class prices twice in this lineage (Opus 4.1 → Opus 4.5, then Opus 5 → Opus 5.5), so treat the three-year column as a ceiling at today’s rates, not a forecast. Batch would halve every figure for asynchronous work; US-only inference would add 10%.
Claude vs OpenAI on the same workload
Using the identical chatbot profile, the raw blended rate (before tokenizer adjustment) for Claude Fable 5.1 is $12.23 per million tokens, against $12.53 for GPT-6 Astra on our OpenAI API pricing page. Opus 5.5 comes in at $4.93 against GPT-6.1 Sol’s $2.47. The two companies’ tokenizers count text differently, so these per-token figures are not a cost-per-document verdict — but the frontier tiers now price within 3% of each other, and the real competition has moved to the mid-tier, where Sol undercuts Opus 5.5 by half per token and Opus 5.5 answers with a cheaper cache.
Claude API Price Change Log: Every Dated Change in 2026
Anthropic rarely announces price moves on their own; they arrive inside model launches. This log is assembled from Anthropic’s launch announcements and deprecations page and is the record we update on every refresh.
| Date | Model / product | Change | Source |
|---|---|---|---|
| 2026-09-30 | Claude Sonnet 4.5 | Deprecated; retirement set for November 30, 2026, replacement claude-sonnet-5-5 | Anthropic deprecations |
| 2026-09-28 | Claude Sonnet 5.5 | Launched at $2 / $10, cache reads $0.20, cache writes $2.50 — same rates as Sonnet 5 | Anthropic |
| 2026-09-22 | Claude Opus 5.5 | Launched at $4 / $20 (−20% vs Opus 5), cache reads $0.20 (−60%), cache writes $5; fast mode $8 / $40 | Anthropic |
| 2026-09-01 | Claude Fable 5.1, Mythos 5.1 | Launched at $10 / $50; cache reads $0.25, down from $1 on Fable 5 | Anthropic |
| 2026-08-05 | Claude Opus 4.1 | Retired | Anthropic deprecations |
| 2026-06-15 | Claude Opus 4, Sonnet 4 | Retired | Anthropic deprecations |
| 2026-06-09 | Claude Mythos Preview | Deprecated | Anthropic deprecations |
| 2026-04-20 | Claude Haiku 3 | Retired | Anthropic deprecations |
| 2026-02-19 | Claude Haiku 3.5, Sonnet 3.7 | Retired | Anthropic deprecations |
The pattern since September is consistent: headline rates fall a little, cache-read rates fall a lot. Fable 5.1 kept its $10 / $50 sticker and cut cache reads 75%. Opus 5.5 cut input and output 20% and cache reads 60%. Sonnet 5.5 held every rate. Anthropic’s pricing strategy is aimed squarely at long-running agents, where cache reads dominate traffic; per-call chatbot traffic sees much smaller gains. For a full release timeline across labs, see our AI model release tracker.
Upcoming dates that will move your Claude bill
- October 15, 2026 — earliest possible retirement of Claude Haiku 4.5, per Anthropic’s deprecations page. Anthropic notifies at least 60 days before retiring a publicly released model, and Haiku 5.5 has been announced for “the coming weeks” without a price.
- November 24, 2026 — earliest possible retirement of Claude Opus 4.5.
- November 30, 2026 — confirmed retirement of Claude Sonnet 4.5. Requests to retired models fail.
What Do Claude’s Built-In Tools and Managed Agents Cost?
| Item | Price | Notes | Source |
|---|---|---|---|
| Web search | $10 per 1,000 searches | Plus search results billed as input tokens | Anthropic |
| Web fetch | No additional charge | Fetched content billed as input tokens | Anthropic |
| Code execution | 1,550 free hours per org per month, then $0.05 per container-hour | Free when used with web search or web fetch | Anthropic |
| Managed Agents session runtime | $0.08 per session-hour | Billed only while status is running | Anthropic |
| Tool-use system prompt | 286 tokens (Opus 5.5, Sonnet 5.5, auto/none); 496 tokens (Haiku 4.5) | Added to input on every request with tools | Anthropic |
| Computer use toolset | ~4,500 input tokens per request | computer_toolset_20260801 defaults | Anthropic |
| Browser use toolset | ~6,600 input tokens per request | browser_toolset_20260801 defaults | Anthropic |
What does a one-hour Claude agent session cost?
Anthropic’s documentation prices a worked example on Opus 5: a one-hour Managed Agents session with 50,000 input and 15,000 output tokens costs $0.705, or $0.525 when 40,000 of those input tokens are cache reads. Axis Intelligence Research re-ran the same session at Opus 5.5 rates:
| Line item | Opus 5.5, no caching | Opus 5.5, 40K cache reads |
|---|---|---|
| Input tokens | 50,000 × $4 / 1M = $0.20 | 10,000 × $4 / 1M = $0.04 |
| Cache reads | — | 40,000 × $0.20 / 1M = $0.008 |
| Output tokens | 15,000 × $20 / 1M = $0.30 | $0.30 |
| Session runtime | 1 hour × $0.08 = $0.08 | $0.08 |
| Total | $0.58 | $0.428 |
That is 18% cheaper than Opus 5 uncached ($0.705) and 18.5% cheaper cached ($0.525). Note the output line: at $0.30 it is 70% of the cached session’s cost. On agent work, the lever left after caching is how much the model writes — which is why Anthropic’s effort settings now matter as much as model choice.
The browser-use toolset deserves a warning of its own. At roughly 6,600 input tokens per request before any page content, an agent making 1,000 browser-tool calls on Opus 5.5 spends about $26 on toolset definitions alone ($4 × 6.6M / 1M) unless those definitions are cached.
What Moves Your Claude API Bill (Beyond the Rate Card)
Cache hit rate. On Opus 5.5 a cache read costs 5% of fresh input; on Fable 5.1, 2.5%. A 5-minute cache write pays for itself after one read; a 1-hour write after two, per Anthropic’s prompt caching guide. Current models cache prompts as short as 512 tokens; Haiku 4.5 needs 4,096. Below the minimum, caching silently does nothing — no error is returned.
The tokenizer. The roughly 30% token inflation on Claude 4.7-and-later models is a real cost line. Measure it on your own content with Anthropic’s token counting endpoint before signing off a migration budget.
Output share and effort. Output costs 5× input on every current Claude model, and thinking is billed as output. Effort level changes how long the model reasons; Anthropic’s Sonnet 5.5 launch post says Sonnet 5.5 at Low or Medium effort beat Sonnet 5’s best benchmark score at about a tenth of the cost per task.
Data residency. US-only inference adds 10% across every token category on Claude 4.6-and-later models, per Anthropic’s data residency documentation.
Where you buy. On Amazon Bedrock and Google Cloud, the cloud provider sets and invoices the price; Anthropic says regional and multi-region endpoints there carry a 10% premium over global endpoints. On Claude Platform on AWS and Microsoft Foundry, Anthropic bills its own standard rates converted to Claude Consumption Units at $0.01 each.
Which Claude Model Should You Pay For?
Analysis by Sarah Mitchell, AI analyst, Axis Intelligence Research.
The rate card says Fable 5.1 is 2.5× Opus 5.5. Anthropic’s own launch post says Opus 5.5 performs at Fable 5.1’s level on most work. Taken together, that is the first time Anthropic has effectively told customers its top-priced model is a specialist tool. The benchmark tables back the cost case more than the capability case — Anthropic itself notes that at this level, benchmark margins have become a less reliable guide to real-world differences. So run your own eval, and price it in TAWR, not stickers.
A defensible default for October 2026:
- Haiku 4.5 for classification, extraction and routing — but build the swap to Haiku 5.5 into your plan now, because its retirement window opens October 15.
- Sonnet 5.5 for well-scoped coding, documents and high-volume chat. Expect savings over Sonnet 4.6 nearer 13% than 33% on price alone; anything beyond that has to come from fewer tokens per task.
- Opus 5.5 for agents and long, cache-heavy sessions. Its $0.20 cache read makes it the strongest price-to-capability point in the lineup for agentic work.
- Fable 5.1 for the hard tail where an Opus failure costs more than the 2.5× premium.
- Batch for anything no one is waiting on.
A price table tells you what a token costs. Only your test set tells you how many tokens a correct answer costs.
Claude API Pricing FAQ
Is Claude Opus 5.5 cheaper than Claude Opus 5?
Yes. Opus 5.5 lists at $4 / $20 per million input/output tokens against Opus 5’s $5 / $25, and its cache reads cost $0.20 versus $0.50, per Anthropic. On a coding-agent workload, Axis Intelligence Research’s TAWR puts Opus 5.5 23% below Opus 5.
Why does my Claude bill rise after migrating to a newer, cheaper model?
Claude 4.7-and-later models use a tokenizer that produces about 30% more tokens for the same text, per Anthropic. A model with a lower per-token rate can still cost more per document. Measure with the token counting endpoint before migrating.
Does the 1M-token context window cost extra on Claude?
No. Anthropic bills Claude 4.6-and-later models at standard per-token rates across the full 1M-token window, with caching and Batch discounts applying throughout. Haiku 4.5’s window is 200K tokens.
Can I combine the Batch API discount with prompt caching?
Yes. Anthropic’s pricing documentation says caching multipliers stack with the Batch discount and with data residency. A cached read on Opus 5.5 in Batch therefore costs $0.10 per million tokens.
When will Claude Haiku 4.5 be retired, and what replaces it?
Anthropic lists Haiku 4.5’s tentative retirement as “not sooner than October 15, 2026.” Anthropic has said Claude Haiku 5.5 will join the Claude 5.5 family in the coming weeks; it had not published Haiku 5.5 pricing as of October 1, 2026.
Is Claude cheaper on Amazon Bedrock or Google Cloud than on the Claude API?
It depends on the provider’s rate card, which AWS and Google set and invoice themselves. Anthropic states that regional and multi-region endpoints on those platforms cost 10% more than global endpoints for Claude 4.5-and-later models.
Is there a free tier for the Claude API?
Anthropic says new users receive a small amount of free credits to test the API. Production use is billed monthly on actual usage, in US dollars.
How much does Claude web search cost inside an agent?
$10 per 1,000 searches, plus the retrieved results billed as input tokens on later turns, per Anthropic. An agent running five searches per task spends $0.05 on search fees before token costs.
Methodology
Price collection. Every list price on this page was read from Anthropic’s pricing documentation on October 1, 2026. Model IDs and context windows come from the models overview; lifecycle dates from the model deprecations page; launch-day price changes from Anthropic’s announcements for Opus 5.5, Sonnet 5.5 and Fable 5.1. No third-party price tracker was used as a source.
Tokenizer-Adjusted Workload Rate (TAWR). TAWR = Σ(token-type share × list price) × k. Workload shares are Axis Intelligence Research assumptions. The tokenizer factor k = 1.3 applies Anthropic’s published “approximately 30%” figure uniformly to Claude 4.7-and-later models (Fable 5.1, Opus 5.5, Sonnet 5.5, Opus 5, Sonnet 5); k = 1.0 for Haiku 4.5, Sonnet 4.6 and Opus 4.1. Anthropic states the real inflation varies by content, so TAWR is an estimate to be recalibrated on your own traffic. TAWR covers token charges only; tool fees, session runtime and residency uplifts are excluded. Monthly and three-year figures hold October 1, 2026 prices constant.
Agent-session recalculation. Inputs mirror Anthropic’s published Opus 5 worked example, repriced at Opus 5.5 rates.
About This Dataset
The dataset behind this page — list prices by model and token type, Batch and fast-mode rates, modifiers, tool prices, lifecycle dates, the dated change log, and every TAWR and cost figure — is available as claude-api-pricing.csv under CC BY 4.0. Every row carries its source URL, retrieval date, a primary-source flag and a flag for Axis-calculated values.
Cite this page:
- APA: Axis Intelligence Research. (2026, October 1). Claude API pricing 2026: Every model, the rate-card change log, and the token tax nobody prices in. Axis Intelligence. https://axis-intelligence.com/claude-api-pricing/
- MLA: Axis Intelligence Research. “Claude API Pricing 2026: Every Model, the Rate-Card Change Log, and the Token Tax Nobody Prices In.” Axis Intelligence, 1 Oct. 2026, axis-intelligence.com/claude-api-pricing/.
- Chicago: Axis Intelligence Research. “Claude API Pricing 2026: Every Model, the Rate-Card Change Log, and the Token Tax Nobody Prices In.” Axis Intelligence, October 1, 2026. https://axis-intelligence.com/claude-api-pricing/.
Axis Intelligence has no commercial relationship with Anthropic. Pricing verdicts are independent.
Related Axis research: OpenAI API pricing 2026 · Claude AI statistics · AI inference cost statistics · AI model release tracker · Claude Opus 4.8 release analysis