Laboratory
Entity that develops or publishes the model and maintains its technical and safety documentation.
VERIFIED ORGANISATION
Anthropic
Explore the documented models, published evaluations and primary sources associated with this organisation.
01 / MODELS
Verified catalogue
KNOWLEDGE AND CODE
Claude Opus 5
Model oriented to code and knowledge work, with controls that allow effort to be adjusted to each task.
16 SEP 2026↗ LONG-RUNNING REASONINGClaude Fable 5.1
Long-running reasoning and agents.
16 SEP 2026↗ COST/CAPABILITY BALANCEClaude Sonnet 5
Coding, research, and professional tool-based workflows.
16 SEP 2026↗ HIGH EFFICIENCYClaude Haiku 4.5
High volume, subagents, and structured extraction.
16 SEP 2026↗ PREVIOUS FRONTIER · AGENTSClaude Opus 4.5
Previous Opus version focused on code, agents, and computer use, retained for migration and historical comparison.
18 SEP 2026↗ AGENTS AND COMPUTER USEClaude Sonnet 4.5
Previous Sonnet generation for code, agents, and interface automation, with official computer use results.
18 SEP 2026↗ AGENTIC CODING AND KNOWLEDGE WORKClaude Opus 5.5
Next-generation Opus for agentic coding, computer use, and extended professional work, with a one-million-token context.
29 SEP 2026↗ BALANCED INTELLIGENCE AND SPEEDClaude Sonnet 5.5
Sonnet update for everyday tasks, code, and documents, with improved speed and a million tokens of context.
29 SEP 2026↗02 / EVALUATIONS
Published results
Figures are shown with the context reported by their source. A provider result is not an independent comparison and does not replace your own evaluation.
| Model | Benchmark | Result | Metric |
|---|---|---|---|
| Claude Opus 5 | Terminal-Bench Science 0.1 | 29,0 % | Accuracy |
| Claude Opus 5 | CursorBench 3.2 | 70,0 % | Accuracy |
| Claude Opus 5 | AutomationBench | 26,9 % | Accuracy |
| Claude Fable 5.1 | Terminal-Bench Science 0.1 | 52,6 % | Accuracy |
| Claude Fable 5.1 | CursorBench 3.2 | 73,4 % | Accuracy |
| Claude Fable 5.1 | Humanity's Last Exam | 65,0 % | With tools |
| Claude Sonnet 5 | OSWorld-Verified | Curve by effort | Success and cost |
| Claude Sonnet 5 | BrowseComp | Curve by effort | Accuracy and cost |
| Claude Haiku 4.5 | SWE-bench Verified | 73,3 % | Resolved |
| Claude Haiku 4.5 | Terminal-Bench | 41,75 % | With reasoning |
| Claude Opus 4.5 | Evaluaciones Claude Opus 4.5 | Published | Code, agents, and computer |
| Claude Sonnet 4.5 | OSWorld | 61,4 % | Computer tasks solved |
| Claude Opus 5.5 | Terminal-Bench 4.0 | 66,4 % | Accuracy |
| Claude Opus 5.5 | GDPval-AA v2.1 | 1.846 | Elo |
| Claude Sonnet 5.5 | Terminal-Bench 4.0 | 70,6 % | Accuracy |
03 / METHOD
How to read the directory
An organization, a product, and a model are not the same unit. Inferama separates them to avoid attributing capabilities or commercial terms to the wrong item.
Model
Identifiable version with limits, modalities, and behavior that may change between releases.
Access channel
API, application, associated cloud, or commercial plan; each channel may have different pricing, retention, and limits.
Source and date
Every claim must retain the official page consulted and the verification date.
ANALYSIS
Related analysis
Claude Sonnet 4.5 with tools: calculate the cost of a task, not just a call
A template for estimating tool-workflow costs by round, separating input and output tokens from additional charges, and checking the calculation against usage logs.
28 Sep 2026 ↗
Claude Opus 4.5 for Long Tasks: Calculate the Cost per Completed Job
A price per million tokens is not enough to budget a multi-step workflow. Learn how to add up input, output, cache, retries, and review, and set limits before you run it.
25 Sep 2026 ↗
Claude Sonnet 5: how to measure the impact of its new tokenizer before migrating
Anthropic says the same text may produce more tokens with Sonnet 5 than with Sonnet 4.6. A test using your own corpus can show whether that reduces usable context or changes cost per task in a specific integration.
23 Sep 2026 ↗
Claude Sonnet 4.5 in the life sciences: what its results show and how to revalidate them
Sonnet 4.5’s scores on biological research tasks justify internal evaluation, not a conclusion about its experimental reliability. What Protocol QA, LAB-Bench, and BixBench measure, what they do not, and how to design a controlled retrospective test.
23 Sep 2026 ↗
Claude Haiku 4.5: How to Calculate Cost per Usable Response
The price per call does not, by itself, show how much an accepted response costs. This guide provides a reproducible formula for adding input and output tokens, retries, and review, with three illustrative scenarios for Claude Haiku 4.5.
23 Sep 2026 ↗
Claude Opus 5 for Desktop Tasks: What Evidence Is Needed Before Giving It Interface Control?
A benchmark score does not prove that a model can operate an application safely. We propose a bounded evaluation to measure success, errors, recovery, latency, and the need for supervision.
23 Sep 2026 ↗04 / SOURCES
Traceability
TECHNICAL SHEETClaude models and limitsplatform.claude.com↗LAUNCHOfficial presentation of Claude Opus 5www.anthropic.com↗TRANSPARENCYClaude Opus 5 transparency reportwww.anthropic.com↗STATEStatus and calendar of Claude modelsdocs.anthropic.com↗BENCHMARKTerminal-Bench Science 0.1www.anthropic.com↗PRIMARY SOURCEOfficial documentationwww.anthropic.com↗LAUNCHOfficial publicationwww.anthropic.com↗LAUNCHOfficial publicationwww.anthropic.com↗TECHNICAL SHEETOfficial release of Claude Opus 4.5www.anthropic.com↗BENCHMARKEvaluaciones Claude Opus 4.5assets.anthropic.com↗TECHNICAL SHEETOfficial release of Claude Sonnet 4.5www.anthropic.com↗TECHNICAL SHEETFicha oficial del modeloplatform.claude.com↗LAUNCHAnuncio o catálogo oficialwww.anthropic.com↗TECHNICAL SHEETFicha oficial del modeloplatform.claude.com↗LAUNCHAnuncio o catálogo oficialwww.anthropic.com↗