Claude Fable 5.1 and GPT-5.6 Luna Take Different AI Paths

For its first CTV campaign, Jennifer Aniston’s DTC haircare brand LolaVie had a few non-negotiables. The campaign had to be simple. It had to demonstrate measurable impact. And it had to be full-funnel.

LolaVie used Roku Ads Manager to test and optimize creatives — reaching millions of potential customers at all stages of their purchase journeys. Roku Ads Manager helped the brand convey LolaVie’s playful voice while helping drive omnichannel sales across both ecommerce and retail touchpoints.

The campaign included an Action Ad overlay that let viewers shop directly from their TVs by clicking OK on their Roku remote. This guided them to the website to buy LolaVie products.

Discover how Roku Ads Manager helped LolaVie drive big sales and customer growth with self-serve TV ads.

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more.

For organizations choosing an AI model in 2026, benchmark scores tell only part of the story. Cost, access, safety controls, provenance rules, and intended workload now matter almost as much as raw model capability.

Anthropic’s Claude Fable 5.1, released on September 1, and OpenAI’s GPT-5.6 Luna represent two different approaches. Fable 5.1 targets demanding reasoning and long-running agentic work. GPT-5.6 Luna is designed primarily for lower-cost, high-volume workloads. Comparing them therefore requires more than asking which model scores higher.

The pricing gap shows the difference immediately.

Anthropic charges $10 per million input tokens and $50 per million output tokens for Fable 5.1. Its important pricing change is cached context. Cache reads now cost $0.25 per million tokens, a 75% reduction from Fable 5. Anthropic estimates this lowers typical workload costs by around 25%. For context-heavy agentic workloads, it estimates reductions approaching 45%.

OpenAI positions GPT-5.6 Luna much lower in the cost structure. After an 80% price reduction announced on July 30, Luna costs $0.20 per million input tokens, $0.02 for cached input, and $1.20 per million output tokens. It supports a 1.05 million-token context window and up to 128,000 output tokens.

This makes the models economically different products. Fable 5.1 is priced for work where model capability may justify considerably higher inference costs. Luna is designed for workloads where millions of repeated classifications, document operations, agent steps, or routine interactions can make token economics central to system design.

Scientific research is where Anthropic is placing much of the Fable 5.1 argument.

On Terminal-Bench-Science 0.1, Anthropic reports Fable 5.1 scoring 52.6%, compared with 24.7% for Fable 5. The same Anthropic evaluation reports GPT-5.6 Sol at 22.4%. Fable 5.1 also reached 55.8% on Terminal-Bench 4.0, while the less-restricted Mythos 5.1 version reached 60.9%.

The 52.6% result deserves attention because it is more than twice Fable 5’s score under Anthropic’s test setup. It should still be treated as a vendor-reported benchmark. Anthropic notes a standard error of roughly 3.5 to 4.5 percentage points per model on Terminal-Bench-Science.

A direct Fable 5.1 versus GPT-5.6 Luna scientific comparison is harder.

OpenAI publishes Luna results across its own science evaluations. GPT-5.6 Luna scores 10.8% on GeneBench Pro, 51.2% on LifeSciBench, and 30.4% on OpenAI’s internal MedChemBench evaluation. It also scores 92.3% on GPQA Diamond. These benchmarks measure different tasks from Terminal-Bench-Science, so the numbers should not be placed beside Anthropic’s 52.6% score as though they measure the same capability.

This distinction matters. Fable 5.1 is intended for difficult research and long-horizon agentic tasks. OpenAI describes Luna as the cost-sensitive tier of the GPT-5.6 family. For complex professional work, OpenAI instead positions GPT-5.6 Sol as its flagship model.

The more unusual part of Anthropic’s release concerns AI-generated text provenance.

Anthropic says models released after August 2, 2026 include an invisible text watermark following its commitments under the EU Code of Practice on Transparency of AI-Generated Content. The watermark does not contain user or conversation information. Detecting it requires Anthropic’s detection system.

That detection API is currently in private preview. Eligible users include regulators, law enforcement bodies, media organizations, fact-checkers, independent researchers, educational organizations, EU civil society groups, and some enterprises with related compliance obligations. Anthropic says access will expand, but it has not provided a public timetable.

This creates an interesting regulatory structure. The text contains a provenance signal, but ordinary readers cannot currently verify that signal themselves.

The issue comes directly from Europe’s changing rules.

Article 50 transparency requirements under the EU AI Act began applying on August 2, 2026. The European Commission says providers covered by these rules must make AI-generated or manipulated content machine-readable so its origin can be detected. Certain systems placed on the market before that date receive a limited transition period until December 2, 2026 for marking and detection requirements.

OpenAI has also endorsed the EU Code of Practice on Transparency of AI-Generated Content. Its current public provenance system, however, is more developed for images and audio than text.

OpenAI Verify can inspect supported images and audio for signals including C2PA metadata and SynthID watermarks. OpenAI says it is working to expand provenance measures across modalities, including text. Its public documentation currently describes text support as a goal rather than an existing public text-verification capability.

That creates a notable difference between the two systems today.

Anthropic has introduced text watermarking for Fable 5.1, but verification remains restricted. OpenAI provides broader public verification for supported image and audio outputs, while text provenance is still being developed.

The comparison therefore does not produce one universal winner.

Fable 5.1 is aimed at expensive, difficult work where scientific research, coding, long-running agents, and deeper reasoning justify greater inference costs. GPT-5.6 Luna approaches the market from the opposite direction. It reduces the cost of running reasoning and tool-enabled models across large workloads.

The more important development may be happening outside benchmark charts. AI companies are increasingly designing model releases around regulation, provenance, access controls, and workload economics.

Model intelligence is still improving. The infrastructure deciding who can use that intelligence, what it costs, and whether its output can later be identified is becoming part of the model itself.