Tech Meridian← ALL MODELS
PROMY MERIDIAN

ANTHROPIC · MODEL RELEASE TRACKER

Claude Opus 4.5

Claude Opus 4.5 is a specific Opus-series model from Anthropic. Anthropic notified API customers that it will retrofit Claude Opus 4.5 with a text watermark starting November 16, 2026; the watermark is reported to apply only when sampling parameters (temperature, top_p, top_k) are left at their defaults and the rollout on Amazon Bedrock may take a few extra days.

CURRENT SNAPSHOT3/5 DIMENSIONS WITH DATA

The dimensions that change the decision.

PRICING

Not established from the available sources.

CONTEXT WINDOW

Not established from the available sources.

MODALITIES
Visual/multimodal capability (participated in photo-based benchmark)Claude Opus 4.5 participated in a benchmark that evaluates models by comparing photographs of IKEA assembly against instructions and identifying mistakes, indicating the model has visual/multimodal capabilities.
BENCHMARKS
Included in AISI/EvalEval public evaluation releaseEvaluation results for Claude Opus 4.5 were included in the UK AI Security Institute (AISI) / EvalEval public release (Evaluation Cards) accompanying the paper 'How Inference Compute Shapes Frontier LLM Evaluation', covering five benchmarks.
Furniture Assembly Benchmark (FAB) score (Nov 2025)In November 2025, Claude Opus 4.5 achieved a 28% score on the Furniture Assembly Benchmark (FAB); at that time it was reported as the best-performing model on that benchmark.
Pass@1 (vs ReAct)Pass@1 remains comparable to ReAct when using Claude Opus 4.5.
LoCoMo LLM-judge accuracy (QEMem evaluation)QEMem with Claude Opus 4.5 achieved 91.43% LLM-judge accuracy on LoCoMo, outperforming the strongest evaluated baseline (MemR^3) by 7.6 percentage points and NEMORI by 10.7 points under a shared memory store.
Included in AI2 'Old Ideas, Novel Problems' novelty-judge benchmarkClaude Opus 4.5 was one of six LLM judges evaluated in the AI2 'Old Ideas, Novel Problems' study and benchmark.
AVAILABILITY
Amazon Bedrock rollout timingAnthropic warned the rollout of the watermark on Amazon Bedrock may take a few additional days relative to the announced date.

VERIFIABLE FACTS

Every value stays attached to a source and date.

BENCHMARKS · Included in AISI/EvalEval public evaluation releaseDEVELOPER CLAIM

Evaluation results for Claude Opus 4.5 were included in the UK AI Security Institute (AISI) / EvalEval public release (Evaluation Cards) accompanying the paper 'How Inference Compute Shapes Frontier LLM Evaluation', covering five benchmarks.

LIMITATIONS · Novelty-evaluation judgments vary strongly with small changes to evaluation setupDEVELOPER CLAIM

The study reported that seemingly small changes to prompts and the evaluation setup produced dramatically different novelty judgments across the six evaluated LLM judges, a group that included Claude Opus 4.5.

SAFETY · Anthropic's claims about the watermark's propertiesDEVELOPER CLAIM

Anthropic describes the watermark as an 'imperceptible pattern in word choice' that does not change meaning, quality, or readability; it reportedly adds no extra characters or tokens, does not affect pricing or request/response formats, 'barely slows' output speed, and 'contains no information about the user, their organization, or their conversations with Claude.'

WHAT CHANGED

Stored passport versions, without reconstructed history.

Passport updated13 facts
AVAILABILITY · Amazon Bedrock rollout timing+ Anthropic warned the rollout of the watermark on Amazon Bedrock may take a few additional days relative to the announced date.
SAFETY · Text watermark rollout date+ Anthropic will apply text watermarking to Claude Opus 4.5 starting November 16, 2026.
SAFETY · Anthropic's claims about the watermark's properties+ Anthropic describes the watermark as an 'imperceptible pattern in word choice' that does not change meaning, quality, or readability; it reportedly adds no extra characters or tokens, does not affect pricing or request/response formats, 'barely slows' output speed, and 'contains no information about the user, their organization, or their conversations with Claude.'
LIMITATIONS · Watermark applies only at default sampling parameters+ Anthropic says the watermark on Claude Opus 4.5 only applies when temperature, top_p, and top_k are set to their default values; changing any one of these sampling parameters can avoid the watermark.
Passport updated9 facts
BENCHMARKS · Included in AI2 'Old Ideas, Novel Problems' novelty-judge benchmark+ Claude Opus 4.5 was one of six LLM judges evaluated in the AI2 'Old Ideas, Novel Problems' study and benchmark.
LIMITATIONS · Novelty-evaluation judgments vary strongly with small changes to evaluation setup+ The study reported that seemingly small changes to prompts and the evaluation setup produced dramatically different novelty judgments across the six evaluated LLM judges, a group that included Claude Opus 4.5.
Passport updated7 facts
BENCHMARKS · LoCoMo LLM-judge accuracy (QEMem evaluation)+ QEMem with Claude Opus 4.5 achieved 91.43% LLM-judge accuracy on LoCoMo, outperforming the strongest evaluated baseline (MemR^3) by 7.6 percentage points and NEMORI by 10.7 points under a shared memory store.
Passport updated6 facts
LIMITATIONS · Completion control trade-off+ Completion control can trade earlier termination for stronger grounding when used with Claude Opus 4.5.
CAPABILITIES · Evidence-complete and constraint-satisfying trajectories+ With Claude Opus 4.5, DeReAct produces more evidence-complete and constraint-satisfying trajectories.
BENCHMARKS · Pass@1 (vs ReAct)+ Pass@1 remains comparable to ReAct when using Claude Opus 4.5.
Passport created3 facts