Claude Haiku 4.5 vs Qwen3-235B-A22B

Best cell per row highlighted. Null means undisclosed — never counted as zero.

Anthropic
Claude Haiku 4.5GA
8.4
AI Panel
Alibaba Cloud
Qwen3-235B-A22BGA
8.5
AI Panel

Identity & lifecycle

ProviderAnthropicAlibaba Cloud
Family / tierClaude 4 · HaikuQwen3 · Large
StatusGAGA
Released2025-10-152025-04-29
Knowledge cutoff2025-022024-10

Architecture & context

Context window200K131K
Max output tokens64K33K
Input modalitiestext, imagetext
Reasoning modeoptionaloptional
Open weightsNoYes

Pricing (per Mtok)

Input$1$0.2
Output$5$0.6
Cached input$0.1—
Batch input$0.5—
Free tierYesYes

Speed

Output speed (tok/s)101.868
Time to first token (s)0.860.78
Latency tierfastmedium

Trust & deployment

Trains on inputsNoNo
LicenseProprietaryApache-2.0
CloudsBedrock, Vertex AI, Azure AI FoundryGCP
ComplianceSOC2, HIPAA, GDPR, ISO27001—

AI Panel scoring

Unified score8.48.5
Decision Maker8.58
Domain Strategist8.58.5
Finance Lead109.5
Domain Practitioner8.59
Power User87.5
Skeptic7.57
Value score9.89.5

Benchmarks

MMLU-Pro7682.8
GPQA Diamond67.270
AIME 2025—81.5
SWE-bench Verified73.3—
LiveCodeBench—70.7
Terminal-Bench41.75—
LMArena Elo14151423
LMArena Coding Elo14821472

Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.