Generalist / documents / analysis / agents

Claude (Anthropic)

Fable 5.1 from 09/01/2026 at Fable 5 pricing; Opus 5 frontier at half the rate; Sonnet 5 default, 1M context

Claude is Anthropic's model family in three tiers: Sonnet 5 for daily agentic work, Opus 5 for frontier tasks at Opus pricing, and Fable 5.1 for the hardest multi-step work. All three have a 1M-token context, up to 128k output and adaptive thinking on by default.

Sonnet 5 (06/30/2026) is the default on Free and Pro and in Claude Code. Opus 5 (07/24/2026) comes close to Fable in many tasks at half the price and is the default on Max. Fable 5.1 (09/01/2026) replaces Fable 5 at the same price but with cache reads 75% cheaper, which cuts the cost of typical workloads by about 25%.

On Max and Team Premium, Fable fits within 50% of the weekly limit; on Pro and Team Standard it runs on usage credits only. The models are available in Claude.ai, Claude Code, the API, Cursor, Amazon Bedrock, Vertex AI and Microsoft Foundry.

  • Verified2026-09-03
  • Models covered6
  • Sources14
  • Deployment channels5
Update 07/24/2026 Claude Opus 5 vs Fable 5 - what changes for production deployments Read analysis Update 07/14/2026 Fable 5 vs the rest: what the strongest coding model really costs Read analysis

Purchase decision (when to choose / when to avoid)

Choose if...

  • You need agents, code and research at good quality-to-price - Sonnet 5 by default.
  • You have frontier tasks (code, knowledge work, agents) and want near-Fable quality at Opus pricing - Claude Opus 5 (from 07/24/2026; API: claude-opus-5).
  • You have frontier tasks and accept higher token cost - Fable 5.1 (from 09/01/2026; Max/Team Premium: up to 50% weekly limits; Pro/Team Standard: usage credits at API rates; API cache reads $0.25 per 1M tok.).
  • You have long documents/contracts and want high-quality analysis (1M context).
  • You build API integrations (claude-sonnet-5) and care about stability and agentic safety.
  • You need enterprise options (e.g. data residency) and transparent materials (system cards).

Avoid if...

  • You have very high volume and token cost is critical - consider cheaper models (e.g. Haiku) or open-weight.
  • You need open-weight and full weight control (on-prem without vendor lock-in).
  • You require Zero Data Retention on Fable 5.1 or Fable 5 - 30-day retention applies; ZDR only with Anthropic approval or once EFS ships (announced for fall 2026).
  • You migrate from Sonnet 4.6 without recalculating cost - the new tokenizer uses more tokens for the same text.

Cost in practice (scenarios)

1-3 people (marketing/ops) - quick start

Typically: Pro/Team plan (Sonnet 5 default) + occasional API.

  • copy + emails + document analysis
  • a few longer files per week
API - daily agentic work (Sonnet 5)

$2/$10 per 1M tok. (standard rate since 08/10/2026; planned increase to $3/$15 canceled). Same text ≈1.0-1.35× more tokens vs old tokenizer.

  • agents, code, research
  • effort level affects cost and quality
Opus 5 - frontier tasks (plans / API / Cursor)

From 07/24/2026: $5/$25 per 1M tok. (same as Opus 4.8); Fast Mode $10/$50. Default on Max, top on Pro.

  • code, knowledge work, multi-step agents
  • near-Fable quality at half the Fable price in many tasks
  • API: automatic fallbacks when safety classifiers flag requests
Fable 5.1 - frontier tasks (plans / API / Cursor)

From 09/01/2026: Max/Team Premium: up to 50% of weekly limits at no extra cost; Pro/Team Standard: usage credits at $10/$50 per 1M tok. Cache reads $0.25 (Fable 5: $1), so typical workloads come out about 25% cheaper and long agentic sessions up to 45% cheaper. In Cursor billed at API rates from the other-models pool.

  • large refactors, catalog migrations, multi-step research
  • cyber and bio prompts route to Opus 4.8 or Opus 5 and are not billed at Fable rates
  • billed from API pool in Cursor, not Auto+Composer
Automations (API) - high volume

Cost depends on context length; with long context it grows non-linearly.

  • RAG / document analysis
  • cache and context compression matter
These are estimates/scenarios (not an invoice). Actual cost depends on context length, number of users, limits and retention policies.

Deployment / data / enterprise

Deployment channels

  • Claude (app) - Sonnet 5 default on Free/Pro; Opus 5 default on Max, top on Pro (from 07/24/2026)
  • Claude API - claude-sonnet-5, claude-opus-5, claude-fable-5-1 (Fable 5 still available); integrations and automations
  • Claude Code - Sonnet 5, Opus 5, Fable 5.1 (version 2.1.255+) and Fable 5
  • Cursor - Opus 5, Fable 5.1 and Fable 5 (billed from the other-models pool, not Auto; Anthropic retention policy acceptance required)
  • Amazon Bedrock / Vertex AI / Microsoft Foundry (depending on region and offering)

Data policy

Training on data
API/Enterprise: not by default (check terms in sources).
Retention
Default API retention is time-limited; Fable 5.1/Fable 5 (and Mythos): mandatory 30-day retention (safety classifiers), data not used for training; ZDR only with Anthropic approval, zero-retention EFS announced for fall 2026.
Data residency
Data residency options available (e.g. inference_geo) - depending on plan/service.
In practice: for sensitive data choose API/Enterprise + retention policies + DPIA.

Enterprise readiness

Admin
Account and policy management in team/enterprise plans.
Best choice when compliance and control matter more than lowest cost.

Best use cases

  • daily agentic work (code, research, automation) at strong quality-to-price - Claude Sonnet 5
  • long document analysis (reports, contracts, specifications) - context up to 1M tokens
  • projects where safety and transparency are a priority (system cards, cyber safeguards, retention policies).
  • frontier tasks at Opus pricing (code, knowledge work, agents) - Claude Opus 5 (from 07/24/2026).
  • hardest multi-step tasks (large refactors, catalog migrations, deep research) - Claude Fable 5.1 (from 09/01/2026; Fable 5 still in the API).

Strengths

  • Sonnet 5: clear gains in reasoning, tool use, coding and knowledge work vs Sonnet 4.6; in some tasks close to Opus 5.
  • Opus 5 (from 07/24/2026): close to Fable 5 in many domains at half the price; strongest Opus knowledge work and coding; Fast Mode ~2.5× faster at $10/$50.
  • 1M context, mature API and data residency; default model on Claude.ai (Free/Pro) and Claude Code.
  • Sonnet 5 pricing: $2/$10 per 1M tok. (standard rate since 08/10/2026; planned increase to $3/$15 canceled); Opus 5: $5/$25; Fable 5: $10/$50.
  • Fable 5.1 (from 09/01/2026): same price as Fable 5, cache reads $0.25 per 1M tok. (75% cheaper); typical workloads about 25% cheaper, agentic up to 45%; Terminal-Bench-Science 52.6% (Fable 5: 24.7%), CursorBench 3.2: 73.4%.
  • Fable 5.1 on plans: Max/Team Premium up to 50% of weekly limits; Pro/Team Standard: usage credits; Claude Code 2.1.255+, Cursor, Bedrock, Vertex AI and Foundry from launch day.

Weaknesses / risks

  • No weights (vendor lock-in); cost grows non-linearly with long context and higher effort levels.
  • New tokenizer = more tokens per prompt; when migrating from Sonnet 4.6 compare effective cost, not just list price.
  • Fable 5.1 and Fable 5 require 30-day retention (safety classifiers); ZDR only with Anthropic approval, and zero-retention EFS is announced for fall 2026. Cyber safeguards enabled by default on Sonnet 5.
  • Fable 5.1 API: forced tool_choice (any/tool) returns 400, thinking blocks are bound to the model and conversation, no Priority Tier - migrating from Fable 5 needs a harness test.

Current models (examples)

  • Claude Sonnet 5 (06/30/2026) - default Sonnet, agentic, 1M context, API: claude-sonnet-5.
  • Claude Opus 5 (07/24/2026) - Opus flagship, coding/agents/knowledge work, thinking on by default, Fast Mode, 1M context, API: claude-opus-5; default on Max, top on Pro.
  • Claude Opus 4.8 (05/28/2026) - Opus 5 predecessor; still available (API: claude-opus-4-8).
  • Claude Sonnet 4.6 (02/17/2026) - Sonnet 5 predecessor; still available but Sonnet 5 is the upgrade.
  • Claude Fable 5.1 (09/01/2026) - Fable 5 successor in the Mythos tier; API: claude-fable-5-1, 1M context, 128k output, $10/$50, cache reads $0.25; Max/Team Premium: up to 50% of weekly limits; Pro/Team Standard: usage credits; Claude Code 2.1.255+, Cursor, Bedrock, Vertex AI, Foundry.
  • Claude Fable 5 (06/09/2026) - Fable 5.1 predecessor; still in the API (claude-fable-5, cache reads $1); same plan rules as Fable 5.1.

Alternatives (if this model doesn't fit)

OP
OpenAI
wider ecosystem and API tooling
GE
Gemini
GCP/Vertex and multimodality in Google ecosystem
MI
Mistral
EU + self-host options (depending on model)