Claude Haiku 4.5 vs Devstral Medium for Writing
In our testing Claude Haiku 4.5 is the better choice for Writing, winning by 1.0 point (3.5 vs 2.5). Haiku 4.5 outperforms Devstral Medium on creative_problem_solving (4 vs 2), persona_consistency (5 vs 3), long_context (5 vs 4), faithfulness (5 vs 4) and tool_calling (5 vs 3), all important for blog posts, marketing copy, and multi-section content. Devstral Medium is significantly cheaper (input/output cost ratio 2.5x in favor of Devstral) but scores lower on the core creative and persona dimensions that matter for Writing.
anthropic
Claude Haiku 4.5
Benchmark Scores
External Benchmarks
Pricing
Input
$1.00/MTok
Output
$5.00/MTok
modelpicker.net
mistral
Devstral Medium
Benchmark Scores
External Benchmarks
Pricing
Input
$0.400/MTok
Output
$2.00/MTok
modelpicker.net
Task Analysis
Writing (blog posts, marketing copy, content creation) demands: creative ideation, concise constrained rewriting, consistent tone/persona, faithfulness to briefs, and long-context handling for multi-section drafts. Because no external benchmark for Writing is included in the payload, our task score is the primary signal: Claude Haiku 4.5 scores 3.5 for Writing vs Devstral Medium's 2.5 in our tests. Supporting internal metrics explain that gap: creative_problem_solving (4 vs 2) drives idea quality and novelty; persona_consistency (5 vs 3) affects brand voice fidelity; long_context (5 vs 4) helps multi-section drafts and series; constrained_rewriting is tied (3 vs 3) so both handle strict-length edits similarly; faithfulness (5 vs 4) reduces brief drift. Cost and parameter differences are secondary trade-offs: Haiku 4.5 has a larger 200,000-token context window and explicit max_output_tokens (64,000), while Devstral Medium’s context is 131,072 tokens and max_output_tokens is not specified in the payload.
Practical Examples
Long-form branded series: Choose Claude Haiku 4.5 — its long_context 5 vs 4 and persona_consistency 5 vs 3 mean it keeps tone and references across chapters. Campaign ideation and hooks: Claude Haiku 4.5 shines thanks to creative_problem_solving 4 vs 2 — expect more non-obvious, feasible ideas. Tight-length ad copy and headlines: Both models tie on constrained_rewriting (3 vs 3), so either can compress to strict limits, but Haiku’s stronger persona and faithfulness reduce tone drift. High-volume short-form production where cost dominates: Devstral Medium is better value — input/output costs are $0.40/$2.00 per mtok vs Claude Haiku 4.5’s $1.00/$5.00 per mtok (Haiku is ~2.5x more costly per mtok). Structured templates and JSON outputs: Both score 4 on structured_output, so both will meet schema-driven content workflows equally well.
Bottom Line
For Writing, choose Claude Haiku 4.5 if you need higher creative ideation, a consistent brand/persona across long documents, stronger faithfulness to briefs, or advanced tool-calling in workflow. Choose Devstral Medium if your priority is lower per-mtok cost and acceptable baseline writing quality for high-volume short-form deliverables.
How We Test
We test every model against our 12-benchmark suite covering tool calling, agentic planning, creative problem solving, safety calibration, and more. Each test is scored 1–5 by an LLM judge. Read our full methodology.