Claude Models Explained: All 32, Opus to Fable 5.1

32 models. That's how many Claude versions Anthropic has released since March 2023. Keeping track of what each model does, which ones are still active, and what they cost is no longer straightforward.

I've used every generation of Claude since version 1.0, from the early 9K context window days to today's 1M-token models such as Fable 5.1 and Mythos 5.1. The progress has been remarkable.

In this article, I'll break down all 32 Claude models, compare their features, pricing, and availability, and give you a clear recommendation for which one to pick.

What Are Claude Models?

Claude models are the AI models behind Anthropic's chatbot. What sets them apart from GPT or Gemini?

A few things:

• Constitutional AI: Claude is trained with a set of principles that ensure responses are helpful, harmless, and honest. More details in Anthropic's Constitutional AI research.
• Large Context Windows: Modern Claude models can process up to 1 million tokens, including Sonnet 5.5, Opus 5.5, Fable 5.1, Mythos 5.1, Fable 5, Sonnet 5, Opus 5, Opus 4.8, Opus 4.7, and the 4.6 models. Every model from Claude 3 onward handles at least 200,000 tokens
• Multimodal Capabilities: From Claude 3 onwards, models can understand and analyze both text and images.
• Hybrid Reasoning: The latest models offer both fast responses and deeper "thinking" for complex problems.

Anthropic has deliberately chosen a different strategy than OpenAI: instead of focusing on maximum performance, safety and control of the AI are the priority.

Before we look at the individual models, here are the key milestones of Claude's evolution from 2023 to today.

Comparison of the 24 Centrally Tracked Claude Models

The table covers all 24 models whose technical specifications, pricing, and status are maintained in this site's central model database. The eight early Claude 1, Claude 2, and Instant variants are covered in the historical sections below.

Claude Fable 5.1 and Claude Mythos 5.1 are Anthropic's new models above the Opus class. Both use the same underlying model. Fable 5.1 is the generally available version with standard safeguards. Mythos 5.1 is the less restricted version for selected organizations.

Shared specifications:

• 1 million token context window and up to 128,000 output tokens
• June 2026 knowledge cutoff as the documented reliable cutoff
• Adaptive thinking is always enabled, with effort set to high by default
• API pricing: $10 per million input tokens and $50 per million output tokens
• Prompt caching: $12.50 for a five-minute write, $20 for a one-hour write, and $0.25 for cache reads
• API IDs: claude-fable-5-1 and claude-mythos-5-1

Fable 5.1 and Mythos 5.1 compared. Claude Fable 5.1 is generally available through the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. Anthropic recommends it for long-running agents, advanced coding, multistep research, and work with documents, spreadsheets, and presentations.

Claude Mythos 5.1 shares Fable 5.1's model, specifications, and pricing, but uses fewer restrictive safeguards. Access remains invitation-only through Project Glasswing. Fable 5.1 is the available version for standard API projects.

What improved over Claude 5? Anthropic reports the largest gains on long-running agentic work. In its announcement, Fable 5.1 scores 55.8% on Terminal-Bench 4.0 and Mythos 5.1 scores 60.9%, compared with 42.0% for Fable 5. On Terminal-Bench-Science 0.1, Fable rises from 24.7% to 52.6%. These are vendor-reported results and should not be mixed with independent benchmark cohorts that use different setups.

Developers also get per-message effort control, turn-scoped system messages, and readable progress updates. Fable 5.1 can emit content provenance information. Migration details matter: forced tool calls can now return an error, older models cannot read the new thinking blocks, and editing earlier messages invalidates existing thinking blocks.

My practical recommendation is straightforward. Pick Fable 5.1 for the most demanding coding, research, and agent workloads. Claude Opus 5.5 is now the much cheaper alternative. Mythos 5.1 is not a standard end-user upgrade. It is a specialized option for organizations with Glasswing access.

Claude Fable 5 and Claude Mythos 5 launched on June 9, 2026 as a new tier above the Opus class that Anthropic calls "Mythos-class". The two models share the same underlying system and the same specs. The difference is the safety classifiers.

Key Features:

• 1 million token context window and 128,000 output tokens
• Adaptive thinking as the only reasoning mode, controlled via the effort parameter
• Safety classifiers on Fable 5: classified refusals fall back to another Claude model (Opus 4.8 by default) and the refusal response is not billed
• 30-day data retention (zero data retention not available, Covered Models)
• Raw chain-of-thought is never returned, only a summary via `thinking.display`
• Pricing: $10 per 1M input tokens, $50 per 1M output tokens (double the Opus 4.8 rate)

Fable 5 vs. Mythos 5: the difference. Claude Fable 5 is the openly available model with active safety classifiers. It is generally available via the Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, and Claude Platform on AWS.

Claude Mythos 5 is the same model without those classifiers. Access runs through Project Glasswing only, for vetted customers in cybersecurity, biomedical research, and critical infrastructure. Customers without Glasswing access use Fable 5.

Brief suspension in June, back online since July 1. Days after the launch, a US export control directive paused access to both Fable 5 and Mythos 5 after security researchers found a jailbreak. Access was restored on July 1, 2026. For subscription plans (Pro, Max, Team, seat-based Enterprise), Fable 5 was included at no extra cost between June 9 and June 22; usage credits were planned to apply from June 23 onward.

Claude Sonnet 5.5 succeeded Claude Sonnet 5 on September 28, 2026 as the newest Sonnet model and is the second model of the Claude 5.5 family after Opus 5.5. Anthropic calls it "a clear upgrade over Claude Sonnet 5" and positions it as a faster, lower-cost complement to Opus 5.5. Per Anthropic, it generates outputs more than 30% faster than Sonnet 5, and in Anthropic's testing it costs up to 30% less per task than its predecessor.

Key Features:

• 1 million token context window and 128,000 output tokens (up to 300,000 via the batch API beta)
• June 2026 knowledge cutoff as the documented reliable cutoff
• Input as text, images, or PDFs, with text output
• Adaptive thinking is on by default. Effort can be set to "low," "medium," "high," "xhigh," or "max," and defaults to "high" in the API and "medium" in Claude Code and the Claude apps
• 81.3% on SWE-Bench Pro and 80.1% on OSWorld 2.1 (Anthropic's system card, measured at max effort, versus 63.2% and 57.0% for Sonnet 5)
• 1,844 Elo on GDPval-AA, against 1,846 for Opus 5.5. On SWE-Bench Pro, Opus 5.5 is well ahead at 89.9%
• Pricing: $2 per million input tokens and $10 per million output tokens, the same as Sonnet 5. Cache reads cost $0.20, a 5-minute cache write $2.50 (1 hour: $4), and the batch API takes 50% off
• API IDs: claude-sonnet-5-5 (Claude API), anthropic.claude-sonnet-5-5 (Amazon Bedrock)

Sonnet 5.5 is the faster, lower-cost complement to Opus 5.5. Anthropic sees its strengths in well-scoped everyday tasks, bug fixing, and polished documents, slides, and spreadsheets. As a curiosity, Anthropic notes that Sonnet 5.5 is the first Sonnet model to beat Pokémon Red working only from screenshots.

The price stays at $2/$10, while speed and benchmark scores rise noticeably. On complex, open-ended work that demands sustained judgment, Anthropic says Opus 5.5 remains clearly stronger. These are vendor figures, so they are not directly interchangeable with independent benchmark cohorts from other test setups.

Claude Sonnet 5 remains active and available. Anthropic has announced no retirement, and the earliest possible date would be June 30, 2027. Anthropic has also said that Claude Haiku 5.5 will join the family in the coming weeks.

You can use Claude Sonnet 5.5 via the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. It also runs in Claude.ai on the web and in the iOS and Android apps, and GitHub Copilot is rolling it out gradually. In Claude Code (version 2.1.284 or later), the "sonnet" alias now points to Sonnet 5.5 on the Anthropic API, while the default model there stays Opus 5.5.

Claude Sonnet 5 became the new default model for Free and Pro users in the Claude apps on June 30, 2026, replacing Claude 4.6 Sonnet as the recommended Sonnet model. Anthropic says it rivals Opus 4.8 at a lower cost.

Key Features:

• 1 million token context window and 128,000 output tokens, and the default in the Claude apps at launch
• Adaptive thinking with an effort parameter that defaults to "high"
• New tokenizer (the same one used by Opus 4.7): about 30% more tokens for the same text than older Sonnet models
• No priority tier available
• Knowledge Cutoff: January 2026
• Pricing: $2 per million input tokens and $10 per million output tokens. Anthropic canceled the previously scheduled increase to $3/$15

At launch, Anthropic made Sonnet 5 the default for the Free and Pro plans, which made it the model most Claude users saw day to day. For the price of a Sonnet model, you get a 1-million-token context window that used to be reserved for the Opus models.

Availability: Claude Sonnet 5 is available via the Claude API, Claude.ai Web Interface, and the Claude iOS and Android apps.

Claude Opus 5.5 succeeded Claude Opus 5 on September 22, 2026 as the newest Opus model. Anthropic calls it "a major step up from Opus 5" and its new leading model. Per Anthropic, Opus 5.5 reaches Fable 5.1-level performance on most work while costing 40% less to run than Opus 5.

Key Features:

• 1 million token context window and 128,000 output tokens (up to 300,000 via the batch API beta)
• June 2026 knowledge cutoff
• Adaptive thinking is always on and can no longer be disabled, with effort defaulting to "medium" (Opus 5 defaulted to "high")
• 66.4% on Terminal-Bench 4.0 and 81.8% on OSWorld 2.0 (Anthropic's own figures, versus 52.3% and 74.0% for Opus 5)
• Separate fast mode: lower latency, currently a research preview available only via the API plus Claude Code and the Claude Platform
• Pricing: $4 per million input tokens, $20 per million output tokens (cheaper than Opus 5's $5/$25). Cache reads cost just $0.20 (0.05x the base input price), batch API: $2/$10
• API IDs: claude-opus-5-5 (Claude API), anthropic.claude-opus-5-5 (Amazon Bedrock)

The core story is again price-to-performance: Anthropic says Opus 5.5 costs 40% less to run than Opus 5 while reaching Fable 5.1-level performance on most work. Anthropic names agentic coding, computer use, and knowledge work as its strengths. Unlike Opus 5, thinking can no longer be turned off on Opus 5.5, but the default effort drops from "high" to "medium," which makes everyday answers cheaper and faster.

Claude Opus 5 remains active and available; Anthropic has not announced a retirement date for it. Availability: Claude Opus 5.5 is available via the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS.

Claude Opus 5 succeeded Claude 4.8 Opus in July 2026 as the recommended Opus model for complex agentic coding and enterprise work. Anthropic described it at launch as close to Claude Fable 5 at half the price. Even after Fable 5.1, Opus 5 remains the cheaper choice for demanding workloads.

Key Features:

• 1 million token context window and 128,000 output tokens (up to 300,000 via the batch beta)
• Knowledge cutoff May 2026
• At launch: default model on Claude Max and the strongest available model on Claude Pro at that time
• Leads Frontier-Bench v0.1 and surpasses Fable 5 on the OSWorld 2.0 benchmark at roughly a third of the cost
• CursorBench 3.2: within 0.5% of Fable 5
• Pricing: $5 per 1M input tokens, $25 per 1M output tokens (half the price of Fable 5, the same as the now-legacy Opus 4.8). Fast mode: double the base price. Batch API: $2.50/$12.50
• Availability at launch: Claude API, Claude.ai web interface, and the Claude iOS and Android apps. Opus 5 became the default on Claude Max and the strongest model on Claude Pro at that time. An API release does not establish an automatic model change in those products, so Fable 5.1's API launch does not prove a new plan assignment

The practical core is the price-to-performance ratio: Opus 5 comes close to Fable 5's capabilities, according to Anthropic, at half the price. Compared with Opus 4.8, Anthropic describes it as "greatly improved performance for the same cost." On ARC-AGI 3, Anthropic reports roughly triple the score of the next-best model, and about 1.5x on Zapier's AutomationBench. It gains 10.2 percentage points over Opus 4.8 on chemistry tasks and 7.7 points on protein-related tasks. On cybersecurity tasks, Opus 5 still trails Claude Mythos 5. Anthropic has not published an official SWE-bench Verified score for Opus 5.

Claude 4.8 Opus was Anthropic's recommended model for agentic coding and enterprise workloads from May to July 2026, before being succeeded by Claude Opus 5. It's now a legacy model but remains available. The jump over Opus 4.7 was smaller than the one from 4.6 to 4.7, but it was real. The biggest gain was once again in coding: SWE-bench Verified rose to 88.6%, up from 87.6% on 4.7. The harder SWE-bench Pro climbed to 69.2%, up from 64.3%.

Key Features:

• 1 million token context window and 128,000 output tokens, same as Opus 4.7
• Dynamic workflows: Claude can spin up hundreds of parallel subagents in a single session to break down very large tasks
• Effort control: you decide how much effort Claude puts into a task (the "high" tier is the default)
• Cheaper fast mode: runs at 2.5x the speed and is three times cheaper than on previous models
• 88.6% on SWE-bench Verified, 69.2% on SWE-bench Pro, 93.6% on GPQA Diamond
• Pricing: $5 per 1M input tokens, $25 per 1M output tokens, unchanged from Opus 4.7
• Availability: Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, Claude apps

The most practically useful improvement was reliability. Anthropic reported that Opus 4.8 was around four times less likely than its predecessor to let flaws in its own code pass unremarked. On pure knowledge it landed roughly level with 4.7, and on GPQA Diamond it even dipped slightly (93.6% vs. 94.2%). The gain was clearly in agentic work, not raw knowledge.

Claude 4.7 Opus was Anthropic's flagship from April to May 2026 and remains active and in use. It built directly on Opus 4.6 and took a clear leap forward in coding: 87.6% on SWE-bench Verified (up from 80.8% on 4.6) and 64.3% on the harder SWE-bench Pro.

Key Features:

• 1 million token context window (now standard, no longer beta)
• 128,000 output tokens
• New tokenizer: The same text now splits into 1.0 to 1.35x more tokens, so real cost can rise slightly at the same per-token price
• "xhigh" reasoning mode: New effort tier between "high" and "max" for finer control over thinking depth
• Image resolution up to 3.75 MP, over 3x prior Claude models; long edge up to 2,576 pixels
• 87.6% on SWE-bench Verified, 94.2% on GPQA Diamond, 78% on OSWorld-Verified, 92.4% on MMLU
• Pricing: $5 per 1M input tokens, $25 per 1M output tokens, same as 4.6 but without the tiered surcharge above 200K
• Availability: Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, Claude apps

Claude 4.6 Opus was the flagship at release and the first Claude model to offer a 1M-token context window alongside Agent Teams for parallel task coordination. It has since been succeeded by Claude 4.7 Opus but remains active and in use.

Key Features:

• 1 million token context window
• Agent Teams for parallel task execution
• 80.8% on SWE-bench Verified, 91.3% on GPQA Diamond
• 200,000 token context window with up to 64,000 output tokens
• Multimodal capabilities: Processes text, images, and documents

Claude 4.6 Sonnet continues the Sonnet tradition of offering excellent performance at a balanced price point. It replaced Claude 4.5 Sonnet as the recommended model for most professional applications and has since been succeeded by Claude Sonnet 5 (June 2026) and Claude Sonnet 5.5 (September 2026).

Key Features:

• Enhanced reasoning and coding capabilities over Claude 4.5 Sonnet
• Improved instruction following compared to Claude 4.5 Sonnet
• 200,000 token context window with up to 64,000 output tokens
• Multimodal capabilities: Processes text, images, and documents
• Best balance of performance and cost for most use cases

Claude 4.5 Haiku is the latest Haiku model and marks a significant advancement in the cost-effective, fast AI category. It's the first Haiku model in version 4.5 and offers an impressive combination of speed, low cost, and surprisingly high performance.

Key Features:

• Significantly improved performance: Surpasses Claude 3.5 Haiku in all benchmarks at similar costs
• Fastest response times: Optimized for high-throughput applications with minimal latency
• Enhanced coding capabilities: Strong performance on programming tasks for a Haiku model
• 200,000 token context window with up to 64,000 output tokens
• Multimodal capabilities: Processes text, images, and documents
• Knowledge Cutoff: January 2025, Training Data through July 2025
• API Pricing: $1 per 1 million input tokens, $5 per 1 million output tokens
• Model String: claude-haiku-4-5-20251001
• Safety Level: ASL-2 (AI Safety Level 2 Standard)

What makes Claude 4.5 Haiku special?

Claude 4.5 Haiku fills an important gap in Anthropic's portfolio: It offers significantly higher performance than its predecessor Claude 3.5 Haiku while maintaining the speed and low costs that Haiku models are known for. This makes it the perfect choice for applications requiring thousands or millions of API calls per day.

Particularly impressive is the coding performance: While earlier Haiku models were primarily intended for simple text tasks, Claude 4.5 Haiku can deliver surprisingly good results even on medium to complex programming tasks, at a fraction of the cost of Sonnet or Opus models.

Ideal Use Cases:

• Chatbots and customer service: Fast, precise responses at low cost
• Content moderation: High-throughput analysis of user-generated content
• Simple automation: Code snippets, formatting, data extraction
• Real-time applications: Where minimal latency is critical
• Batch processing: Processing large volumes of documents or text

Availability: Claude 4.5 Haiku is available via the Claude API, Claude.ai Web Interface, Claude iOS and Android Apps, Amazon Bedrock, and Google Cloud Vertex AI.

Claude 4.5 Sonnet (September 2025) was a significant breakthrough for Anthropic. It combines world-class coding performance with the attractive price of the Sonnet series: For the first time, a Sonnet model surpassed the Opus variants in the most important benchmarks. It has since been succeeded by Claude 4.6 Sonnet (February 2026).

Key Features:

• State-of-the-art coding performance: Leading in SWE-bench Verified Benchmark, surpasses even Claude 4.1 Opus
• Impressive focus ability: Can stay focused for 30+ hours on complex, multi-step tasks
• OSWorld leadership: 61.4% performance in OSWorld benchmark (vs. 42.2% for Claude 4 Sonnet four months earlier)
• 200,000 token context window with up to 64,000 output tokens
• Multimodal capabilities: Processes text, images, and documents
• Knowledge Cutoff: January 2025, Training Data through July 2025
• API Pricing: $3 per 1 million input tokens, $15 per 1 million output tokens
• Model String: claude-sonnet-4-5-20250929
• Safety Level: ASL-3 (AI Safety Level 3 Standard)

What makes Claude 4.5 Sonnet special?

Claude 4.5 Sonnet marks a paradigm shift in Anthropic's model strategy. While previously Opus models offered the absolute peak performance, Anthropic has managed to develop a model with Claude 4.5 Sonnet that surpasses the Opus variants in most areas at one-fifth of the cost.

The most remarkable capability is "Extended Focus": Claude 4.5 Sonnet can work focused for over 30 hours on complex development tasks without losing quality or coherence. This makes it ideal for large refactoring projects, complex system architectures, and multi-step debugging sessions.

Claude 4.1 Opus was the most advanced Opus model of its generation, building on the Claude 4 family, which represented a quantum leap in AI development. These Hybrid-Reasoning models can switch between two modes: lightning-fast responses for simple questions and extensive "thinking" for complex problems. It has since been succeeded by Claude 4.5 Opus and Claude 4.6 Opus.

Key Features:

• Improved reasoning capabilities compared to Claude 4 Opus
• Strong coding assistant with a 74.5% success rate on SWE-bench Verified
• Extended Hybrid-Reasoning functions with even deeper "thinking"
• Significantly reduced hallucination rate (under 2% in benchmarks)
• API Pricing: $15 per 1 million input tokens, $75 per 1 million output tokens
• 200,000 token context window
• Improved multimodal capabilities, especially for technical diagrams

Claude 4 Opus was the original flagship model of the Claude 4 family, which introduced Hybrid-Reasoning technology.

Key Features:

• Previous best coding assistant with 72.5% success rate in SWE-bench (now surpassed by 4.1 Opus)
• Hybrid-Reasoning: Users can choose between fast and thoughtful responses
• Excellent for complex scientific, mathematical, and programming tasks
• API Pricing: $15 per 1 million input tokens, $75 per 1 million output tokens
• 200,000 token context window
• Multimodal capabilities (text, images, documents)

Claude 4 Sonnet is the balanced version of the Claude 4 family and offers the best balance between performance and cost for most applications.

Key Features:

• Also equipped with Hybrid-Reasoning capabilities
• Better price-performance ratio than Opus
• API Pricing: $3 per 1 million input tokens, $15 per 1 million output tokens
• Ideal for most professional applications
• Same technical foundation as Opus
• 200,000 token context window

Claude 3.7 Sonnet was the first model to introduce Anthropic's new Hybrid-Reasoning technology. It allows users to switch between fast responses and an "extended thinking mode."

Key Features:

• First Claude with controllable thinking time
• Toggle between "Fast" and "Think" mode
• Improved performance on complex reasoning tasks
• 200,000 token context window
• Multimodal input (text and images)
• Test model for later Claude 4 development

Claude 3.5 Haiku is the fastest and most cost-effective model in the Claude 3.5 series, which significantly improved upon the Claude 3 family foundations.

Key Features:

• Fastest and most cost-effective model in the series
• Optimized for high throughput rates
• Impressive quality despite speed
• Perfect for chatbots and simple automation
• Multimodal capabilities at low cost
• 200,000 token context window
• API Pricing: $0.80 per 1 million input tokens, $4 per 1 million output tokens

Claude 3.5 Sonnet was the first model in the improved 3.5 series and offered better performance at similar costs to Claude 3.

Key Features:

• Surpasses Claude 3 Opus in many benchmarks
• Significantly cheaper than Opus with comparable performance
• Improved coding capabilities and visual understanding
• Better workflow integration and tool use
• 200,000 token context window
• Ideal for productive applications
• API Pricing: $3 per 1 million input tokens, $15 per 1 million output tokens

Claude 3 Opus was the most powerful model in the Claude 3 family, which marked a turning point: for the first time, Claude models could process both text and images.

Key Features:

• The most powerful model in the series
• First Claude version with true multimodality
• Excellent for complex analysis and reasoning tasks
• 200,000 token context window
• Discontinued on June 30, 2025 (Retirement on January 5, 2026)

Claude 3 Sonnet was the balanced model in the Claude 3 family, which introduced the Japanese poetry-inspired naming convention.

Key Features:

• Balanced model between performance and cost
• Good all-round capabilities for most applications
• Multimodal capabilities (text and images)
• 200,000 token context window
• Was replaced by Claude 3.5 Sonnet on July 21, 2025

Claude 3 Haiku was the fastest and cheapest model in the Claude 3 family.

Key Features:

• Fastest and cheapest model in the Claude 3 series
• First multimodal Claude version for high-throughput applications
• Discontinued in April 2026
• 200,000 token context window
• Multimodal capabilities (text and images)
• API Pricing: $0.25 per 1 million input tokens, $1.25 per 1 million output tokens

Claude 2.1 was the improved version of Claude 2, the first publicly available Claude.

Key Features:

• 200,000 token context window (twice as large as Claude 2.0)
• Reduced hallucinations compared to Claude 2.0
• Improved accuracy for long documents
• Discontinued

Claude 2.0 was the first publicly available Claude and brought significant improvements over the first generation.

Key Features:

• 100,000 token context window
• Improved coding capabilities
• Better mathematical reasoning
• Available via API and Claude.ai website
• Discontinued

Claude 1.3 was the final version of the first Claude generation, laying the foundation for Anthropic's approach to AI safety.

Key Features:

• Improved safety features through Constitutional AI
• Better handling of complex instructions
• Only available via API
• Discontinued

Claude 1.2 was an intermediate version of the first Claude generation.

Key Features:

• Incremental improvements over Claude 1.0
• Better text understanding
• Discontinued

Claude 1.0 was the very first model from Anthropic and laid the foundation for all later developments.

Key Features:

• First model with Constitutional AI training
• 9,000 token context window
• Focus on safety and helpfulness
• Foundation for all later developments
• Discontinued

Which Claude Model Should You Choose?

The choice of the right Claude model depends on your specific requirements:

For Developers and Programmers

Claude Sonnet 5.5 is the sensible default for most developers. It combines a 1M-token context window with the $2/$10 API price of Sonnet 5, faster output, and, per Anthropic, a clear step up on coding benchmarks.

Pick Claude Fable 5.1 for the hardest long-running coding and agent workloads. Claude Opus 5.5 is the much cheaper alternative, and Anthropic says it stays clearly ahead of Sonnet 5.5 on complex, open-ended work. Mythos 5.1 is relevant only if your organization has Project Glasswing access.

For Businesses and Professional Applications

Claude Sonnet 5.5 offers the best price-performance ratio for most business applications. It brings a 1 million token context window, keeps the $2/$10 rate of Sonnet 5, and in Anthropic's testing costs up to 30% less per task than its predecessor. That makes it a strong fit for customer service, content creation, and data analysis.

For High-Volume Applications

Claude 4.5 Haiku (October 2025) is the newest and best Haiku model for chatbots, automated responses, and other applications that need fast, cost-effective responses with high quality. It offers significantly better performance than Claude 3.5 Haiku at a similar price point (see the pricing table below).

Price Comparison of All Claude Models

The following table shows a detailed overview of all Claude prices (all figures in $ per million tokens). All active models offer 50% discount when using the Batch API.

Important notes on the price table:

• All prices are in USD per million tokens
• Batch API usage provides 50% discount on all active models
• Prompt caching can reduce costs by up to 90% for repeated content
• Extended thinking tokens in Opus/Sonnet 4+ models are billed at the regular output rate

Claude models in a consistent Vals AI evaluation setup

Models:Fable 54.8 Opus4.7 Opus4.6 Opus4.6 Sonnet4.5 Opus4.5 Sonnet4.1 Opus

Claude Fable 5

Claude 4.8 Opus

Claude 4.7 Opus

Claude 4.6 Opus

Claude 4.6 Sonnet

Claude 4.5 Opus

Claude 4.5 Sonnet

Claude 4.1 Opus

Source: Vals AI, uniform cohort

PNG|SVG

CC BY 4.0

gradually.ai

It gets even more interesting when you plot coding performance against API price and compare Claude with GPT and Gemini.

Price and performance in the same SWE-bench cohort

Ideal: strong + cheap

Anthropic

OpenAI

Google

DeepSeek

Moonshot AI

Efficiency frontier (best price-performance)

Sources: Vals AI, vendor price lists

PNG|SVG

CC BY 4.0

gradually.ai

Conclusion

For most users, Claude Sonnet 5.5 (September 2026) is the right choice. It offers strong everyday and coding performance at a balanced price point, with a 1M token context window and the $2/$10 rate of Sonnet 5.

For high-volume, cost-sensitive applications, Claude 4.5 Haiku offers the best balance of speed, quality, and price. For demanding coding and enterprise work, choose Fable 5.1 when capability matters most or Opus 5.5 when you want to cut the token price by more than half.

The good news: all modern Claude models are excellent. You can't really go wrong. Just pick the one that fits your budget and use case.