TL;DR: Claude and OpenAI no longer fit a simple “which flagship is cheaper” answer. Claude Sonnet 5 is the first Claude production test at $2/$10 per 1M tokens through August 31, then $3/$15. Opus 5 is the premium Claude baseline at $5/$25. OpenAI’s standard short-context GPT-5.6 family spans Luna at $0.20/$1.20, Terra at $2/$12, and Sol at $5/$30.

Anthropic says the June 12 export controls were lifted June 30 and Fable 5 returned globally on native Claude surfaces July 1. Included usage is temporary and plan-specific, cloud re-enablement is rolling out, and Mythos 5 remains restricted to approved organizations. Fable’s restored availability does not make it the default premium lane.

GPT-5.6 Sol, Terra, and Luna have been generally available since July 9. OpenAI cut Luna’s API price by 80% and Terra’s by 20% on July 30; Sol was unchanged. Product access is plan-dependent, while less-restricted cyber work uses Trusted Access. See the current model guide and Luna viability analysis.

Use this page for routing decisions, not rankings. Test before moving production workloads.

Current API Price Anchors

Prices are per 1 million tokens.

LaneInputCached inputOutputBatch input/outputPractical read
Claude Sonnet 5$2 intro; $3 standardverify current cache rate$10 intro; $15 standardverify current pricing docsIntroductory pricing ends Aug 31
Claude Opus 5$5$0.50 cache hits$25$2.50 / $12.50Premium Claude baseline; tune effort
Claude Fable 5$10$1 cache hits$50$5 / $25Restored guarded escalation; not a default lane
GPT-5.6 Sol, short context$5$0.50 cache reads$30$2.50 / $15Generally available flagship
GPT-5.6 Terra, short context$2$0.20 cache reads$12$1 / $6Generally available balanced tier
GPT-5.6 Luna, short context$0.20$0.02 cache reads$1.20$0.10 / $0.60Generally available high-volume tier
GPT-5.5 standard, short context$5$0.50$30$2.50 / $15OpenAI-native premium lane
GPT-5.5 standard, long context$10$1$45$5 / $22.50Long context when OpenAI fit matters
GPT-5.5 Pro, short context$30not listed$180$15 / $90Highest-cost OpenAI Pro lane

GPT-5.6 model pages list a 1.05M context window. Requests above 272K input use long-context rates for the full request: Sol $10/$45, Terra $4/$18, and Luna $0.40/$1.80 input/output. Cache writes cost 1.25 times the applicable input rate.

Cost Sanity Check

For a 100K-input / 20K-output request:

ModelApprox standard cost
Claude Sonnet 5$0.40 introductory / $0.60 standard
GPT-5.6 Luna standard short context$0.044
GPT-5.6 Terra standard short context$0.44
Claude Opus 5$1.00
GPT-5.6 Sol standard short context$1.10
GPT-5.5 standard short context$1.10
GPT-5.5 standard long context$1.90
Claude Fable 5$2.00
GPT-5.5 Pro short context$6.60

Those numbers are only a starting point. Repeated-context workloads can favor OpenAI or Claude cache-hit pricing. Batch workloads halve both providers’ listed input and output token charges. Data residency, priority/fast modes, and tool/runtime charges can change the bill.

When Claude Wins

Choose Claude when:

  • Sonnet 5 solves the task and you want Claude behavior at the lowest current Claude production price.
  • Opus 5 improves architecture, debugging, review, or long-horizon agent work enough to justify $5/$25.
  • You need Claude Code or claude.ai as the first-party workflow.
  • You want to test Fable only after target-route access is confirmed, Opus misses a high-value task, and your data policy allows Fable’s retention terms.

Do not route sensitive workloads to Fable just because a benchmark or launch quote looks strong. Fable can refuse or fall back on guarded domains, and Anthropic documents 30-day retention with no zero-data-retention option for Fable/Mythos.

When OpenAI Wins

Choose the matching OpenAI tier when:

  • Luna clears a bounded, high-volume task’s acceptance bar with controlled retries and review.
  • Terra materially reduces Luna’s failure or cleanup rate.
  • Sol changes the result on hard agentic, terminal, or final-review work.
  • Cached input dominates the bill.
  • OpenAI-native tooling, context, or API behavior fits better than Claude’s route.

GPT-5.5 standard short context is cheaper than Fable on both input and output. GPT-5.5 long context matches Fable input pricing and is cheaper on output. GPT-5.5 Pro is much more expensive than Fable, so use it only when the Pro behavior is measurably worth it.

Decision Matrix

If your workload is…Start withEscalate to
Routine production codingClaude Sonnet 5 or GPT-5.5 standardOpus 5 if review/architecture quality matters
Premium Claude reviewOpus 5 highRaise effort after task-level measurement; use Fable only after access, compliance, and guardrail tests pass
OpenAI-native bounded volumeGPT-5.6 LunaTerra when Luna’s retries or review erase the saving
OpenAI-native hard agentic workGPT-5.6 TerraSol only when it changes the accepted result
Long-context repeated promptsCompare cache-hit pricing on both providersBatch or long-context lanes after a real bill estimate
Cyber, biology, chemistry, or model-development workGuardrail test firstTrusted access or another approved model
Zero-data-retention requirementAvoid Fable/Mythos-class routesUse a model/contract that meets the requirement

What To Test Before Buying

  • Representative prompts from your real workload.
  • Cost per successful task, not cost per token alone.
  • Fallback/refusal logs for Fable-sensitive domains.
  • Cache-hit rate on repeated context.
  • Batch fit for asynchronous workloads.
  • Data retention, data residency, and audit logging requirements.
  • Tooling fit: Claude Code, Codex, cloud platform, or BYOK harness.

Sources


Last updated: August 1, 2026. GPT-5.6 is generally available with separate short- and long-context rates; Opus 5 is the current premium Claude baseline; Fable 5 is restored and Mythos 5 remains restricted. Pricing, included usage, cloud availability, retention, and model terms can change independently.