manual budget
Only provider-documented reasoning controls are shown; missing controls remain unpublished rather than inferred.
MODEL REFERENCE / ANTHROPIC
Claude Haiku 4.5
verified model reference.
Fastest Claude tier for latency-sensitive and high-volume workloads.
200,000context tokens
64,000max output
2026-09-27last verified
PRIMARY-SOURCE VERIFIED
Claude Haiku 4.5 at a glance.
Anthropic documents a 200,000-token context window. Input modalities: text + image. Output: text. Pricing status: Official paid API.
MODEL IDclaude-haiku-4-5-20251001
CONTEXT WINDOW200,000tokens
MAX OUTPUT64,000
KNOWLEDGE CUTOFF2025-02
STANDARD INPUT$1.00per 1 million tokens
STANDARD OUTPUT$5.00per 1 million tokens
CAPABILITY CONTRACT
Published model behavior.
text + image
Input modalities come from the cited official model documentation.
text
Output modality and output-limit claims remain separate so an undocumented token cap is not guessed.
PRICING STATUS
Official paid API.
$1.00 input · $0.10 cached · $5.00 output per 1 million tokens.
SXF preserves the provider-published native billing basis. Only calculator-eligible generative token models enter token-cost arithmetic; specialist units and input-only embeddings remain visible without forced conversion.
EXPLORE
Keep the model in context.
OFFICIAL SOURCES
Verification links.
WHAT CHANGED
Claude Haiku 4.5: what changed.
No post-baseline factual changes recorded for Claude Haiku 4.5 since 2026-09-27.
VERIFIED BASELINE2026-09-27Current facts remain aligned with the SXF ledger.
VERIFIED CHANGE HISTORY
Claude Haiku 4.5 verified history.
Claude Haiku 4.5
Baseline verified: 200,000 context · 64,000 max output · $1.00 input / $0.10 cached / $5.00 output per 1 million tokens
Official evidence ↗Append-only SXF ledger · 1 verified event · chain head b12eeb63a07e…
EVALUATION EVIDENCE
Claude Haiku 4.5 benchmark observations.
independentArtificial Analysis
AA-Briefcase 1.1
614 Eloagentic knowledge work · reasoning reasoning
Evidence ↗independentArtificial Analysis
Artificial Analysis Intelligence Index 4.3.2
17composite intelligence · reasoning reasoning
Evidence ↗independentArtificial Analysis
AA-LCR 1.1
74.0%long context reasoning · reasoning reasoning
Evidence ↗independentArtificial Analysis
AutomationBench-AA
3.0%business automation · reasoning reasoning
Evidence ↗independentArtificial Analysis · under review
CritPt
0.0%frontier science reasoning · reasoning reasoning
Evidence ↗independentArtificial Analysis
GDPval-AA 2.1
735 Eloprofessional knowledge work · reasoning reasoning
Evidence ↗independentArtificial Analysis
Humanity's Last Exam
10.0%multidisciplinary reasoning · reasoning reasoning
Evidence ↗independentArtificial Analysis · under review
SciCode
42.0%scientific coding · reasoning reasoning
Evidence ↗Scores are shown only with their evaluator, benchmark version and model configuration. They are not merged into a universal model score.