Home › Pricing › Anthropic

Anthropic Claude API Pricing (October 2026)

Last updated:

All Anthropic Claude Models — Price per 1M Tokens

Showing 6 current models. .
Relative priceLowestLowerHigherHighestAutomatic log scale across current offers; input, cache and output are graded separately. Color shows cost, not quality.
Showing 6 grouped models from 6 offers · USD per 1M tokens
Last synced:

Sponsored links may earn us a commission at no extra cost to you. Affiliate status never changes model ordering.

All product names, logos, and brands are property of their respective owners and are used for identification purposes only.

Claude morality story, not a model launch: an October 4 HN link recirculates a September 29 feature about Anthropic's constitution. Anthropic announced the current constitution in January; it did not announce a new Claude SKU or token rate with this story. The live table above remains the pricing source. Read the pricing and Labs evidence boundary →

Show HN FIRE calculator cost boundary: the public salary-to-retirement simulator runs locally and has no Anthropic SDK or hosted-model dependency. Claude built the project and Claude Code is an optional plan-editing interface, but the repository publishes no model version, token ledger, or development bill. Claude rates above and paid-plan allowances remain unchanged. See the verified pricing, financial-model risks, license gap, and Labs blocker →

Opus 5.5 in Claude Code: a newly resurfaced developer guide recommends complete tasks with explicit acceptance tests, mid-run steering, and measured effort/fast-mode choices. It announces no new SKU or rate. Direct API pricing above differs from Pro/Max plan allowances; fast mode doubles standard input/output rates to $8/$40 per million. Read the workflow and cost guide →

Claude Opus 5.5 is now current: $4.00 input, $0.20 cache reads, $5.00 five-minute cache writes, and $20.00 output per 1M tokens. That is 20% below Opus 5 for input/output and 60% below it for cache reads. Read the launch analysis →

Samaya earnings study: Fable 5.1 and Opus 5.5 beat a bias-corrected analyst-consensus baseline in Samaya's finance-specific test. This is not an Anthropic price change or a published cost per forecast. The direct Claude rates above remain the official model-only baseline. Read the model-cost and replay boundary →

September 22 availability incident resolved: Anthropic reported major elevated errors across Claude.ai, the API, Claude Code, and Cowork, with customer impact from 00:50 to 02:10 UTC. Anthropic closed the incident at 02:35 UTC. No model, endpoint, lifecycle status, or token rate changed. See the official timeline, cost impact, and recovery checklist →

Fresh Labs check: the exact anthropic/claude-opus-5.5 route scored 49/49 with zero errors for $0.05422 total. GPT-6 Sol also scored 49/49 for $0.015922 on the same deterministic suite. Compare measured task cost →

Claude inside Siri is an unreleased route: private iOS 27 and macOS 27 framework demonstrations show Claude as a Model Delegation extension, but Apple has not opened the entitlement or published a Claude-for-Siri price, allowance, account requirement, or supported-region list. Anthropic's direct API rates above remain the only official Claude cost baseline. See what the code proves, what Apple officially supports, and the Labs blocker →

Pacing-the-frontier regulation debate: Dario Amodei proposed slowing capability growth and Anthropic committed to embedded outside evaluators. David Sacks endorsed voluntary pacing but rejected regulation or antitrust relief as the prerequisite. No Claude model, endpoint, rate, service level, or release date changed; current API prices remain the measurable baseline. See the primary-source positions, pricing impact, and buyer checklist →

US open-weight distillation policy debate: Garry Tan called for regulators to leave room for legitimate American distillation, while Anthropic's official position supports open weights but seeks action against industrial-scale unauthorized extraction. No model, endpoint, or rate changed with the debate; Claude's current prices remain the measurable source-model baseline, not permission to use outputs for training. See the full-cost model, evidence boundary, and buyer checklist →

Which Claude should be your default? A September 12 Ask HN discussion repeatedly separated value defaults such as Sonnet or older Opus routes from premium Fable escalation. That is useful buyer-language evidence, not a benchmark or a price change. See the current cross-provider cost and routing guide →

Cami Clark governance story, not a Claude launch: the September 10 Hacker News submission recirculates an August profile about Dario Amodei's wife and her reported informal influence. Clark is not listed on Anthropic's leadership page, the story introduces no model or API SKU, and Claude pricing is unchanged. Read the sourced governance boundary and buyer checklist →

Fable 5.1 is live: Anthropic kept the $10 input / $50 output headline rate but cut cache reads from Fable 5's $1.00 to $0.25 per 1M tokens. Fable 5 remains available as a legacy route; Mythos 5.1 shares the new rates but is limited to trusted-access programs. The separate 65k-line Go-to-Rust case study used Fable 5 and reports subscription-subsidized usage rather than a replayable API bill. See the $400 rewrite evidence and buyer test plan →

Claude Code Opus 5 Auto mode security watch: a targeted indirect prompt-injection chain reached code execution in 3/5, 3/5, and 4/5 small-sample variants. That proves a bypass of the tested setup, not a universal 60–80% attack rate. Anthropic's own deployed-classifier report records 17% false negatives on 52 real overeager actions and 5.7% on 1,000 synthetic exfiltration attempts. Opus 5 pricing is unchanged at $5 input / $0.50 cached input / $25 output per 1M tokens; Auto mode is a permission feature, not a separate API SKU or sandbox. See the pricing boundary, evidence limits, and containment checklist →

Model Hardware Standard research preview: Anthropic's MHS release is a model-agnostic hardware interoperability specification, not a new Claude model or API SKU. Preview access is application-based, Anthropic published no standalone MHS fee, and current Claude token rates did not change. See the pricing boundary, access status, buyer checklist, and Labs blocker →

College-essay student vote: Claude-family answers placed second in StudyArena's 6,851 eligible blind writing votes at a 31.8% choice rate, behind Gemini at 39.6% and ahead of OpenAI at 29.2%. StudyArena also reports Claude leading assignment planning at 43.2%, but the provider-family aggregation and missing ballot-level dataset prevent an independent replay. Claude Opus 5 remains $5 input / $25 output per 1M tokens. Read the pricing-aware student-vote analysis →

Vomit's separate-LLM Claude 5 rewrite: the GPLv3 Claude Code hook sends Claude's already-generated message through a second OpenAI-compatible model. It has no published fee and recommends local GPT-OSS 20B, but it does not erase Claude output tokens already billed; local compute, latency, and rewrite fidelity remain buyer costs. The project publishes no controlled token-saving benchmark, and Anthropic's model rates did not change. See the cost boundary, risks, and Labs blocker →

Graft's 42% Claude Code token claim: the MIT-licensed context tool reports mean task tokens falling from 8,070 to 4,650 in its own 162-run Sonnet 5 benchmark. Graft has no published subscription or per-token fee, and Anthropic's model rates did not change; optional deep graph enrichment still uses the buyer's provider key. Treat 42% as a vendor-run workload result, not a universal grep discount. See the pricing boundary, benchmark audit, and buyer test →

Spotify Portal's 90% Claude Code context claim: Spotify's Apache-2.0 shunt plugin reported 82–94% less context returned to Claude across three large-file reads by routing files through a Gemini worker. The benchmark estimates tokens as characters divided by four and omits the worker's full usage, Portal cost, repeated trials, and accepted-task quality, so it does not prove a 90% lower total bill. Anthropic's model rates did not change. See the benchmark audit, cost boundary, and paired test plan →

You own Claude outputs, but model-training rights are narrower: Anthropic assigns customers its rights in outputs, subject to the applicable terms. Its current policy permits specialized, non-competing classifiers and tools, while prohibiting competitive general-purpose models, open-ended text generators, using outputs as training targets, and helping third parties do the same. The policy introduced no new model or training surcharge; separately, Anthropic made Sonnet 5's current $2/$10 rate permanent by cancelling its scheduled September increase. See the terms, allowed uses, and buyer checklist →

Claude Code's reported 12×–40× pricing gap: Anthropic did not change its model rates. The new buyer signal compares usage included with eligible Max seats against the counterfactual cost of the same recorded tokens at API list rates. Anthropic's current usage-based Enterprise plan separately bills every token at standard API rates on top of its seat fee, so teams should replay their own model and cache mix before migrating. See the evidence, caveats, and calculator →

How Claude's text watermark works: Anthropic says future Claude models will use a SynthID-Text-style statistical pattern in word selection—not hidden characters or user tracking—and supported PNG, JPG, and SVG files will use separate signed C2PA credentials. The mark should be stronger in long prose and weaker in short, factual, lightly edited, or code-heavy output. Anthropic says it adds no tokens, speed penalty, token-rate change, or marking surcharge; a detection API is coming but has no published access or price yet. See the mechanism, limitations, and developer checklist →

15-day Claude app rewrite buyer watch: Boris Cherny described an internal Claude Tag agent trying to reproduce Anthropic's Electron desktop app in Swift. The experiment was still running after 14–15 days, and no model snapshot, token count, spend, or finished artifact was disclosed. Anthropic's public API prices have not changed. Read what buyers can and cannot conclude →

Opus 5 buyer watch: Anthropic's official rate remains $5.00 input / $25.00 output per 1M tokens, but early Claude Code users report mixed tool and context behavior. Our current Labs run scored 47/49 with zero API errors and higher output use than Opus 4.8. Read the evidence and migration test plan →

New Claude 5 context guidance: Anthropic says it removed more than 80% of Claude Code's system prompt for Opus 5 and Fable 5 without a measurable loss on its coding evaluations. The pricing did not change, but shorter recurring context can reduce input-token cost and latency. Read the Claude 5 context engineering cost guide →

Sonnet 5 launched: Anthropic introduced Claude Sonnet 5 on June 30, 2026. For most production routers, Sonnet 5 is now the default Claude price/performance option below Opus 5 and Fable 5. Read Anthropic's announcement →

Anthropic's Claude models are known for strong reasoning, coding, and safety. Claude Opus 5.5 is Anthropic's recommended choice for most coding and knowledge work, Fable 5.1 is the higher-priced escalation tier, and Sonnet 5 remains the price/performance option. Claude Mythos 5.1 is the vetted trusted-access lane.

Claude pricing is generally higher than OpenAI across comparable tiers, but Anthropic competes on coding performance, agent reliability, long context in supported models, and aggressive prompt-caching discounts.

Current Claude Lineup: Opus 5.5, Fable 5.1, and Mythos 5.1

  • Claude Opus 5.5 ($4.00 input / $0.20 cached / $20.00 output) - Anthropic's recommended default for most coding and knowledge work. It supports 1M context, up to 128K synchronous output, and adaptive thinking.
  • Claude Fable 5.1 ($10.00 input / $0.25 cached / $50.00 output) - The higher-priced escalation tier for demanding reasoning and long-running agentic work that still performs better in a team's own evaluation. It supports 1M context and up to 128K output.
  • Claude Mythos 5.1 ($10.00 input / $0.25 cached / $50.00 output) - The same core model and pricing with a trusted-access safeguard regime for approved cybersecurity and life-sciences work.
  • Claude Fable 5 ($10.00 input / $1.00 cached / $50.00 output) - Legacy route used in the reported 65k-line Go-to-Rust rewrite. Audit the $400 case study →
  • Claude Opus 5 ($5.00 input / $0.50 cached / $25.00 output) - Previous-generation Opus route retained for compatibility and migration comparisons.
  • Claude Sonnet 5 ($2.00 input / $10.00 output) - Best price-to-performance ratio in the Claude lineup. 1M context, up to 128K output, strong for everyday coding and production workflows. Anthropic says the launch rate is now the standard price.

JuliaHub's independent sealed simulation study adds a useful workload-specific check: Fable 5 led the physical-AI score, while GPT-5.6 Sol cost much less per trial. Use that result for engineering-agent routing evidence, not as a universal model ranking.

Claude Haiku (Budget Tier)

Haiku models provide the best value in Anthropic's lineup:

  • Claude Haiku 4.5 ($1.00 input / $5.00 output) — Fast, capable, good for production workloads
  • Claude Haiku 3.5 ($0.80 input / $4.00 output) — Popular mid-tier option, strong quality-to-cost ratio
  • Claude Haiku 3 ($0.25 input / $1.25 output) — Anthropic's cheapest model. Great for classification, extraction, and high-volume tasks.

Prompt Caching Discounts

Anthropic's prompt caching is one of the most aggressive in the industry. Current cache-read discounts range from 90% to 97.5% below base input price:

  • Claude Opus 5.5: $0.20/M cached (vs $4.00 standard input) - 95% savings
  • Claude Fable 5.1 and Mythos 5.1: $0.25/M cached (vs $10.00 standard) - 97.5% savings
  • Legacy Fable 5 and Mythos 5: $1.00/M cached (vs $10.00 standard) - 90% savings
  • Claude Opus 5: $0.50/M cached (vs $5.00 standard) - 90% savings
  • Claude Sonnet 5: $0.20/M cached (vs $2.00 standard input) - 90% savings

Batch API offers an additional 50% discount, making Claude Haiku 3 batch pricing just $0.125/M input — competitive with the cheapest models on the market.

Legacy Models: Still Available, Often Expensive

Earlier Claude models like Fable 5, Mythos 5, Opus 4.8, 4.7, and 4.6 are now previous-generation. Fable 5 keeps the same $10/$50 input/output rate as 5.1 but costs four times as much for cache reads.

How Claude Compares

Claude Opus 5.5 ($4.00 input) is Anthropic's recommended premium default and costs 60% less than Fable 5.1 at equal fresh-input volume. Fable 5.1 remains the higher-priced escalation route and should be reserved for work where a team's own evaluation shows that the extra capability materially changes the result.

The real value proposition for Claude is coding quality, frontier reasoning, agent reliability, and cache economics. Many developers should route everyday work to Sonnet 5, escalate difficult work to Opus 5.5, and reserve Fable 5.1 for jobs where fewer failed agent loops justify the higher base token rate.

For the absolute cheapest option, check DeepSeek's pricing at $0.28/M for their flagship model.

Price History

Only models with a recorded price change are charted here.

Claude Mythos 5.1

Claude Mythos 5

Claude Opus 4

Claude Opus 4.1

Claude Sonnet 4

Claude Haiku 3.5

Price history tracking started April 2026. Flat model charts stay hidden until a price change is detected.
View pricing changelog →

Frequently asked questions

How much does Claude Opus 5.5 cost?

Claude Opus 5.5 costs $4 per 1M input tokens, $0.20 per 1M cache reads, $5 per 1M five-minute cache writes, $8 per 1M one-hour cache writes, and $20 per 1M output tokens. Fast mode costs $8 input and $40 output per 1M.

How much does Claude Fable 5.1 cost per token?

Claude Fable 5.1 costs $10.00 per 1M input tokens, $0.25 per 1M cache reads, and $50.00 per 1M output tokens. Five-minute cache writes cost $12.50 per 1M and one-hour writes cost $20.00 per 1M.

How much does Claude Fable 5 cost per token?

Claude Fable 5 is priced at $10.00 per 1M input tokens and $50.00 per 1M output tokens. Cache hits and refreshes cost $1.00 per 1M, 5-minute cache writes cost $12.50 per 1M, and 1-hour cache writes cost $20.00 per 1M. Anthropic says Fable 5 access was restored on July 1, 2026.

What is Claude Mythos 5.1?

Claude Mythos 5.1 shares Fable 5.1 pricing and core capabilities, but Anthropic limits it to approved trusted-access programs such as Project Glasswing rather than normal self-serve availability.

Does Anthropic have a free tier?

Anthropic does not offer an ongoing free API tier. New accounts receive $5 in trial credits. The Claude.ai consumer app has free usage, but the developer API requires a paid account or trial credit.

How does Claude Fable 5.1 compare to Opus 5.5?

Fable 5.1 costs $10 input and $50 output per 1M, versus $4 and $20 for Opus 5.5. Anthropic recommends Opus 5.5 for most workloads and Fable 5.1 when demanding long-horizon work still benefits from the frontier tier.

How much does Claude Sonnet 5 cost per token?

Claude Sonnet 5 costs $2.00 per 1M input tokens, $0.20 per 1M cache reads, and $10.00 per 1M output tokens. Anthropic says this is now the standard price and the previously scheduled September 1 increase will not occur.

What Claude models does Anthropic offer?

Anthropic's current lineup includes Claude Opus 5.5 for most coding and knowledge work, Fable 5.1 for the hardest long-horizon work, Sonnet 5.5 for price/performance, and Haiku 4.5 for lower-cost production work. Mythos 5.1 is the trusted-access route.

What is Claude's context window?

Claude Opus 5.5, Fable 5.1, Mythos 5.1, Opus 5, and Sonnet 5 support a 1M-token context window. Opus 5.5, Fable 5.1, Opus 5, and Sonnet 5 list up to 128K synchronous output tokens.

How much does Claude prompt caching save?

Fable 5.1 and Mythos 5.1 cache reads cost $0.25 per 1M, 97.5% below their $10 base input rate and 75% below Fable 5 cache reads. Claude Sonnet 5.5 and Opus 5.5 use a 95% cache-read discount ($0.10 and $0.20 per 1M respectively); Sonnet 5 remains at $0.20 per 1M, a 90% discount.

Does the Show HN FIRE calculator require Claude or Claude Code?

No. The calculator runs locally as deterministic Python with Rich and Plotext. Claude was used to build it and Claude Code is an optional customization interface. Paid Claude plans include Claude Code; Console-authenticated usage follows standard API pricing.

Methodology

Pricing sourced from https://www.anthropic.com/pricing on . All prices in USD per 1 million tokens. Raw data: /api/pricing.json. API docs.

Compare All Providers

See how Anthropic stacks up against OpenAI, Google, DeepSeek, and more.

View All Models Token Calculator OpenAI Pricing Voice/TTS Pricing Public API

Looking for voice/TTS API pricing? Compare ElevenLabs, Speechify, OpenAI audio, Google TTS, and Amazon Polly.