Claude Fable 5 vs Qwen2.5-Coder-32B-Instruct vs Codestral (25.08) vs GPT-5.4

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

Anthropic
Claude Fable 5GA
9.5
AI Panel
Alibaba Cloud
Qwen2.5-Coder-32B-InstructGA
8.4
AI Panel
Mistral AI
Codestral (25.08)GA
7.6
AI Panel
OpenAI
GPT-5.4GA
8.7
AI Panel

Identity & lifecycle

ProviderAnthropicAlibaba CloudMistral AIOpenAI
Family / tierClaude 5 · FableQwen2.5-Coder · CoderCodestral · CoderGPT-5 · Medium
StatusGAGAGAGA
Released2026-06-092024-11-122025-07-312026-03-05
Knowledge cutoff2026-012024-082025-052025-08

Architecture & context

Context window1M131K128K1.1M
Max output tokens128K8K16K128K
Input modalitiestext, image, filetexttexttext, image
Reasoning modealwaysnonenoneoptional
Open weightsNoYesNoNo

Pricing (per Mtok)

Input$10$0.08$0.3$2.5
Output$50$0.24$0.9$15
Cached input$1—$0.03$0.25
Batch input$5—$0.15$1.25
Free tierNoYesYesNo

Speed

Output speed (tok/s)———86.3
Time to first token (s)———1.2
Latency tierslowfastfastmedium

Trust & deployment

Trains on inputsNoNoNoNo
LicenseProprietaryApache-2.0ProprietaryProprietary
CloudsBedrock, Vertex AI, Azure AI FoundryGCPBedrock, Azure AI Foundry, Vertex AIAzure OpenAI, Azure AI Foundry
ComplianceSOC2, HIPAA, GDPR, ISO27001—SOC2, ISO27001, GDPRSOC2, GDPR, CCPA, HIPAA

AI Panel scoring

Unified score9.58.47.68.7
Decision Maker9.58.57.59
Domain Strategist9.587.58.7
Finance Lead7989
Domain Practitioner9.59.589
Power User98.57.58.5
Skeptic7.57.578
Value score6.59.58.58.5

Benchmarks

GPQA Diamond———92.8
SWE-bench Verified95———
HumanEval—92.786.6—
LiveCodeBench—31.4——
MMMU———81.2
Terminal-Bench88——75.1
MRCR Long Context———36.6
Humanity's Last Exam———39.8
LMArena Elo15061270—1476
LMArena Coding Elo15521342—1520

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