Anthropic Claude API pricing October 2026 (official rates)

This is Anthropic Claude API pricing as of October 2026: Anthropic's official per-token list price for every current Claude model, side by side with what the same model costs on Kunavo, plus worked cost examples and the levers that cut a Claude bill the most. Claude is the default choice for reasoning, coding and agentic work, and Kunavo undercuts Anthropic's list on every Claude model — 30% off each of them — behind one OpenAI-compatible API. Sonnet 5 sat at Anthropic's list rate until September 24, 2026; the note below has the history.

In one line: Anthropic lists Claude from $1.00 input / $5.00 output per 1M tokens (Claude Haiku 4.5) to $10.00 / $50.00 (Claude Fable 5), and prompt-cache reads bill at 10% of the input rate (2.5% on Claude Fable 5.1, 5% on Claude Opus 5.5).

Rates last verified October 1, 2026. Kunavo's per-token prices below are read live from the model catalog, and the “Anthropic list” column tracks Anthropic's published rate — the official source is anthropic.com/pricing (developer detail at docs.anthropic.com).

Official Anthropic API pricing — Claude Opus, Sonnet and Haiku (October 2026)

Rates are per 1M tokens, in USD, as billed on Kunavo. The “Anthropic list” column is Anthropic's official published rate for the same model, taken from anthropic.com/pricing on the verification date above — so this table is the official price list and the discounted one in a single view.

Live rates always show on the pricing page and each model page. Haiku is the cheap workhorse, Sonnet the balanced default, Opus the heavy reasoner — and the Claude 5 family adds Sonnet 5 (near-Opus coding at Sonnet cost), Opus 5 and Fable 5, Anthropic's most capable model here for frontier reasoning and long-horizon agents. Opus 5.5, released September 22, 2026, is the newest Opus and the first model of Anthropic's Claude 5.5 family.

Claude Opus 5.5 is the value pick. Anthropic lists it at $4.00 / $20.00 per 1M — below Opus 5's $5.00 / $25.00 — and says it "performs at the level of Claude Fable 5.1 on most work". On Kunavo it is $2.80 / $14.00, ~30% under Anthropic's list and below Opus 5's $3.50 / $17.50. A frontier-grade run that costs $0.42 on Fable 5 (50K in / 2K out) costs $0.21 on Opus 5 and $0.168 on Opus 5.5 — thinking is always on for Opus 5.5, so budget its output line on the thinking too. The Opus 5.5 API guide has the model ID, the breaking changes and working code.

Claude Sonnet 5 pricing, and why it changed

Anthropic originally launched claude-sonnet-5 at $2.00 / $10.00 per 1M as an introductory rate, with a rise to $3 / $15 scheduled for September 1, 2026. On August 31, 2026 Anthropic cancelled that increase and made $2.00 / $10.00 the permanent standard price. Kunavo sold it at exactly that price until September 24, 2026, and now charges $1.40 / $7.00 — 30% under list. Beyond the per-token price, Kunavo gives you on Sonnet 5 what it gives on every model: one key and one balance across Claude, GPT and the image/video models, no Anthropic account, no monthly minimum, and no charge for failed calls.

For Opus-tier reasoning, claude-opus-5-5 costs $2.80 / $14.00 on Kunavo, 30% under Opus 5.5's own $4.00 / $20.00 list price. Claude Fable 5 is a straight ~30% saving from the first call: $7.00 / $35.00 against Anthropic's $10.00 / $50.00.

Anthropic API pricing by model — Haiku, Sonnet, Opus and Fable

The same figures broken out per model family, since Anthropic prices each tier separately and most cost questions are really about which tier a workload belongs in. Every rate is per 1M tokens, input / output.

Claude Haiku 4.5 API pricing

Anthropic lists claude-haiku-4-5 at $1.00 / $5.00 — the cheapest model in the Claude family and the one to reach for on high-volume, low-difficulty work: classification, extraction, routing, support replies. On Kunavo it is $0.70 / $3.50, about 30% under list. At that rate an 800-in / 200-out support ticket costs about $0.00126, so 10,000 tickets a day is roughly $12.60/day. Haiku 4.5 has a 200K context window rather than the 1M the rest of the current family carries.

Claude Sonnet API pricing — Sonnet 4.6 and Sonnet 5

The two Sonnet generations no longer carry the same Anthropic list price. claude-sonnet-4-6 — the everyday production workhorse — lists at $3.00 / $15.00 and is $2.10 / $10.50 on Kunavo, about 30% under list. claude-sonnet-5 brings near-Opus coding and agentic quality at Sonnet cost and lists lower, at $2.00 / $10.00 — the introductory rate Anthropic made permanent on August 31, 2026. Kunavo prices it at $1.40 / $7.00, about 30% under list — see the note below. A 6K-in / 500-out RAG answer is about $0.0179 on Sonnet 4.6.

Claude Opus API pricing — Opus 4.7, Opus 5 and Opus 5.5

The Opus tier is the heavy reasoner. Anthropic lists both claude-opus-4-7 and claude-opus-5 at $5.00 / $25.00; Kunavo serves both at $3.50 / $17.50, about 30% under list — the deepest discount in the family, and the reason moving an Opus 4.7 workload to Opus 5 costs nothing extra. A 50K-in / 2K-out long-context analysis on Opus 5 costs about $0.21.

claude-opus-5-5, released September 22, 2026, lists lower than both, at $4.00 / $20.00, and reads its cache at 5% of input instead of 10%. Kunavo serves it at $2.80 / $14.00, about 30% under list; the same 50K-in / 2K-out analysis costs about $0.168.

Claude Fable 5 API pricing

claude-fable-5 is Anthropic's most capable model on Kunavo — frontier reasoning, long-horizon agents, 1M context — and its most expensive. Listed at $10.00 / $50.00 per 1M tokens, double the Opus tier. On Kunavo it is $7.00 / $35.00, about 30% under Anthropic's list, so that saving applies from the first call. The cost math that matters: a 50K-in / 2K-out frontier agent run is about $0.42 on Fable 5 against $0.21 for the same run on Opus 5 — so reserve Fable 5 for work where the Opus tier has actually been tried and fallen short. Thinking is always on for this model, and reasoning tokens bill at the output rate, which makes the output line the one to watch.

How Anthropic API pricing works per token

Anthropic prices the Claude API per token, quoted per 1M tokens and billed on the exact count — there is no per-request fee, no minimum and no subscription. You pay for input tokens (system prompt, context, messages) and output tokens (the completion). Output is billed at roughly 5× the input rate across the family, so the largest lever on cost is how much the model writes. The second lever is prompt caching: stable, repeated context can be cached and billed at a fraction of the normal input rate.

A token is about 4 characters of English, so 1M tokens ≈ 750,000 words. At Kunavo's $10.50 per 1M output on Sonnet 4.6, a 500-token answer costs about $0.00525 in output. Reasoning models add a wrinkle: hidden reasoning tokens are billed at the output rate even though you never see them, which is why an output cap saves more than it looks like it should.

Worked cost examples

Real numbers at Kunavo's rates:

So 10,000 support tickets a day on Haiku is about $12.60/day. The math, runnable:

# Kunavo Claude rates (USD per 1M tokens): (input, output)
RATES = {
    "claude-haiku-4-5":  (0.70, 3.50),
    "claude-sonnet-4-6": (2.10, 10.50),
    "claude-sonnet-5":   (1.40, 7.00),
    "claude-opus-4-7":   (3.50, 17.50),
    "claude-opus-5":     (3.50, 17.50),
    "claude-opus-5-5":   (2.80, 14.00),
    "claude-fable-5":    (7.00, 35.00),
    "claude-fable-5-1":  (7.00, 35.00),
}

def cost(model: str, in_tokens: int, out_tokens: int) -> float:
    i, o = RATES[model]
    return in_tokens / 1_000_000 * i + out_tokens / 1_000_000 * o

print(cost("claude-haiku-4-5", 800, 200))      # support ticket -> $0.00126
print(cost("claude-sonnet-4-6", 6_000, 500))   # RAG answer     -> $0.0179
print(cost("claude-sonnet-5", 10_000, 1_500))  # coding agent   -> $0.025
print(cost("claude-opus-4-7", 50_000, 2_000))  # long analysis  -> $0.21
print(cost("claude-opus-5", 50_000, 2_000))    # Opus 5 analysis-> $0.21
print(cost("claude-opus-5-5", 50_000, 2_000))  # Opus 5.5       -> $0.168
print(cost("claude-fable-5", 50_000, 2_000))   # frontier run   -> $0.42

Prompt caching — the biggest single saving

If your system prompt or retrieved context is large and stable, prompt caching bills cached input at 10% of the input rate (2.5% on Claude Fable 5.1, 5% on Claude Opus 5.5). For a long, reused system prompt that is a 60–90% cut on the input line. It is one extra field on the request — see the Anthropic prompt caching deep dive for the exact pattern and the Messages API reference. What that looks like on a bill, including the three ways a gateway can quietly break your hit rate, is in Claude prompt caching; the Claude token cost calculator has a cached-tokens field so you can price it against your own traffic.

Kunavo pricing and Stripe billing

No subscription, no Anthropic Console billing setup. Top up a balance (Stripe or local payment methods) and calls draw down at the rates above. Pay-as-you-go from a $10 minimum top-up, the balance never expires, and larger top-ups carry bonus credit. The same balance covers Claude, GPT, image, video and audio models — one wallet, one key.

Picking a Claude tier: the cost decision

Match the model to the job rather than defaulting to the most capable one:

See the GPT pricing guide for OpenAI's models, and the cost optimization guide for the difficulty-routing pattern in code.

Claude Code API pricing

Claude Code — Anthropic's agentic coding CLI — runs on these same Claude models, so its cost is ordinary Claude API token cost: the files and tool output it reads count as input tokens, and the edits and explanations it writes count as output tokens. Sonnet 5 ($1.40 / $7.00 per 1M) is the newest near-Opus coder and costs less per token here than Sonnet 4.6 ($2.10 / $10.50), and Opus 4.7 ($3.50 / $17.50) or the frontier Fable 5 ($7.00 / $35.00) handle the hardest refactors. Agentic sessions re-read context each turn, so they are token-heavy — which makes prompt caching the biggest lever, billing cached context at 10% of the input rate (2.5% on Claude Fable 5.1, 5% on Claude Opus 5.5). Point Claude Code (or another agentic coding tool) at Kunavo's Anthropic-compatible Messages endpoint to pay the rates above — 30% under Anthropic's list depending on the model — instead of list price.

Claude Code reads ANTHROPIC_BASE_URL natively, so that redirect is three environment variables and no extra software — the Claude Code setup guide has the exact variables and the credential mistake behind most 401s. If you are still deciding between paying per token and paying for a plan, Claude Code pricing puts the Pro/Max plan fees next to the per-token rates and works through the break-even, is Claude Code free covers what is and isn't included, and getting an API key for Claude Code covers where the key goes. Not installed yet? Install Claude Code has the per-OS command.

FAQ