Claude Sonnet 4.6 vs Devstral Small 1.1

Head-to-head across our 12-test suite (scores from the payload): Sonnet wins 9 tests and ties 3; Devstral wins none. Breakdown (Sonnet score vs Devstral score): strategic_analysis 5 vs 2 (Sonnet — nuanced tradeoff reasoning; ranks tied for 1st of 54), creative_problem_solving 5 vs 2 (Sonnet tied for 1st of 54), tool_calling 5 vs 4 (Sonnet tied for 1st of 54; Devstral rank 18 of 54) — meaning Sonnet is more reliable at selecting functions, sequencing calls, and producing correct arguments. Faithfulness 5 vs 4 (Sonnet tied for 1st of 55) — Sonnet sticks to sources better in our tests. Long_context 5 vs 4 (Sonnet tied for 1st of 55; Devstral rank 38 of 55) — Sonnet performs better on retrieval/consistency across 30K+ token contexts. Safety_calibration 5 vs 2 (Sonnet tied for 1st of 55; Devstral rank 12 of 55) — Sonnet better refuses harmful requests while allowing legitimate ones. Persona_consistency 5 vs 2 (Sonnet tied for 1st of 53) and agentic_planning 5 vs 2 (Sonnet tied for 1st of 54) — Sonnet wins on maintaining personas and decomposing goals. Multilingual 5 vs 4 (Sonnet tied for 1st of 55) — Sonnet gives stronger non-English parity. Ties: structured_output 4 vs 4 (both rank 26 of 54), constrained_rewriting 3 vs 3 (both rank 31 of 53), classification 4 vs 4 (both tied for 1st of 53). External benchmarks (supplementary): Claude Sonnet 4.6 scores 75.2% on SWE-bench Verified (Epoch AI), ranking 4 of 12, and 85.8% on AIME 2025 (Epoch AI), ranking 10 of 23; Devstral has no external scores in the payload. In practice: choose Sonnet when you need top tool-calling accuracy, safety, faithfulness, and long-context reasoning; choose Devstral when cost and throughput dominate and tied areas (structured output/classification) are the primary needs.

Costs per 1,000 tokens (mTok) from the payload: Claude Sonnet 4.6 — input $3, output $15; Devstral Small 1.1 — input $0.10, output $0.30 (priceRatio = 50). Example monthly bills assuming 1,000 mTok per 1M tokens: 1M tokens (1000 mTok) — Claude: $3,000 (all input) to $15,000 (all output); Devstral: $100 to $300. At 10M tokens — Claude: $30,000 to $150,000; Devstral: $1,000 to $3,000. At 100M tokens — Claude: $300,000 to $1,500,000; Devstral: $10,000 to $30,000. If you expect sustained high-volume inference (millions of tokens/month), the cost gap becomes strategic: startups, high-volume API customers, and low-margin production services should prioritize Devstral; teams that require the higher benchmarked capabilities and top safety/faithfulness should budget for Claude.

Choose Claude Sonnet 4.6 if you need best-in-class tool calling, safety calibration, long-context retrieval, agentic planning, and multilingual parity for production, developer tooling, or high-stakes workflows and can absorb higher per-token costs. Choose Devstral Small 1.1 if you need an inexpensive model for high-volume inference, prototypes, or cost-sensitive production where structured output and classification parity (ties) are sufficient and the top-tier safety/agentic/creative performance is not required.