Home › Comparisons › Model comparison Price vs performance of coding models
Each dot is a model that publishes both an API price and a DeepSWE v1.1 score. Higher means a better reported score; further left means cheaper.
Read this before comparing: DeepSWE v1.1 scores are reported by each vendor, with its own agent harness and reasoning effort. They are not strictly comparable with each other, and a gap of a few points is not enough to rank two models.
Open weights All Yes No
Vendor All vendors Alibaba (Qwen) DeepSeek Google MiniMax Moonshot AI xAI Z.ai
10 of the 30 models in the comparison have both a price and a DeepSWE v1.1 score.
Price vs performance of coding models Scatter plot of 10 models: horizontal axis, blended price (3 input tokens per output token) in dollars per million tokens, logarithmic scale; vertical axis, DeepSWE v1.1 score reported by the vendor. The same values are in the table below the chart. $0.2 $0.5 $1 $2 $5 $10 40% 50% 60% 70% 80% Blended 3:1 price ($ per million tokens, log scale) Reported DeepSWE v1.1 score (%) Gemini 4 Argon (Google) · input $2 · output $10 · blended $4 · DeepSWE v1.1: 77.9% (announced by Google) · Open weights: No · tiered or time-limited pricing: base price Gemini 4 Argon DeepSeek V4.1-Flash-1017 (DeepSeek) · input $0.3 · output $1.2 · blended $0.53 · DeepSWE v1.1: 77% (announced by DeepSeek, Mini-SWE-V2 harness) · Open weights: Yes · tiered or time-limited pricing: base price DeepSeek V4.1-Flash-1017 MiniMax M3.1 Flash (MiniMax) · input $0.15 · output $0.6 · blended $0.26 · DeepSWE v1.1: 73.7% (announced by MiniMax) · Open weights: No MiniMax M3.1 Flash Grok 4.7 (xAI) · input $2 · output $6 · blended $3 · DeepSWE v1.1: 71% (high effort, announced by xAI) · Open weights: No Grok 4.7 Kimi K3 (Moonshot AI) · input $3 · output $15 · blended $6 · DeepSWE v1.1: 67.3% (mini-SWE-agent harness (DeepSWE leaderboard), cited by Moonshot) · Open weights: Yes Kimi K3 GLM-5.3 (Z.ai) · input $1.4 · output $4.4 · blended $2.15 · DeepSWE v1.1: 66.9% (announced by Z.ai) · Open weights: Yes GLM-5.3 GLM-5.3-Flash (Z.ai) · input $0.15 · output $0.5 · blended $0.24 · DeepSWE v1.1: 63.4% (announced by Z.ai) · Open weights: Yes GLM-5.3-Flash DeepSeek V4-Pro-0813 (DeepSeek) · input $1.32 · output $3.96 · blended $1.98 · DeepSWE v1.1: 62.7% (DeepSeek Harness (minimal mode), max effort, announced by DeepSeek) · Open weights: Yes · tiered or time-limited pricing: base price DeepSeek V4-Pro-0813 Qwen3.8-Max (Alibaba (Qwen)) · input $2 · output $6 · blended $3 · DeepSWE v1.1: 56.6% (best of Claude Code / mini-SWE-agent, announced by Qwen) · Open weights: Yes Qwen3.8-Max Qwen3.8-27B (Alibaba (Qwen)) · input $0.5 · output $3 · blended $1.12 · DeepSWE v1.1: 42.2% (Claude Code harness, announced by Qwen) · Open weights: Yes Qwen3.8-27B
Hover, tap or tab through the dots to show their details. The table further down lists every value.
No dot selected.
Blended price = (3 × input price + 1 × output price) ÷ 4, in dollars per million tokens, from standard list prices (no caching or discounts). The 3:1 input-to-output ratio is a display convention, not a measurement: a coding agent that reads many files often uses far more input than output. The horizontal axis is logarithmic: the same distance means the same price ratio ($0.50 to $1 just like $5 to $10). For models with tiered or launch pricing, the documented base price is used.
Chart data
Chart data
Model Vendor Input price ($/M tokens) Output price ($/M tokens) Blended 3:1 price ($/M tokens) DeepSWE v1.1 Open weights Pricing conditions Sources Gemini 4 Argon Google $2 $10 $4 77.9%announced by Google No launch price until Nov 30, 2026, then $4/$20tiered or time-limited pricing: base price price — Gemini 4 Argon · score — Gemini 4 Argon DeepSeek V4.1-Flash-1017 DeepSeek $0.3 $1.2 $0.53 77%announced by DeepSeek, Mini-SWE-V2 harness Yes peak hours, cache miss; half price off-peaktiered or time-limited pricing: base price price — DeepSeek V4.1-Flash-1017 · score — DeepSeek V4.1-Flash-1017 MiniMax M3.1 Flash MiniMax $0.15 $0.6 $0.26 73.7%announced by MiniMax No flat rate; $0.015 cached input price — MiniMax M3.1 Flash · score — MiniMax M3.1 Flash Grok 4.7 xAI $2 $6 $3 71%high effort, announced by xAI No — price — Grok 4.7 · score — Grok 4.7 Kimi K3 Moonshot AI $3 $15 $6 67.3%mini-SWE-agent harness (DeepSWE leaderboard), cited by Moonshot Yes cache-miss input; $0.30 cache hit price — Kimi K3 · score — Kimi K3 GLM-5.3 Z.ai $1.4 $4.4 $2.15 66.9%announced by Z.ai Yes — price — GLM-5.3 · score — GLM-5.3 GLM-5.3-Flash Z.ai $0.15 $0.5 $0.24 63.4%announced by Z.ai Yes — price — GLM-5.3-Flash · score — GLM-5.3-Flash DeepSeek V4-Pro-0813 DeepSeek $1.32 $3.96 $1.98 62.7%DeepSeek Harness (minimal mode), max effort, announced by DeepSeek Yes peak hours, cache miss; half price off-peaktiered or time-limited pricing: base price price — DeepSeek V4-Pro-0813 · score — DeepSeek V4-Pro-0813 Qwen3.8-Max Alibaba (Qwen) $2 $6 $3 56.6%best of Claude Code / mini-SWE-agent, announced by Qwen Yes international (Singapore) price price — Qwen3.8-Max · score — Qwen3.8-Max Qwen3.8-27B Alibaba (Qwen) $0.5 $3 $1.12 42.2%Claude Code harness, announced by Qwen Yes international (Singapore) price price — Qwen3.8-27B · score — Qwen3.8-27B
Not on the chart: models without a vendor-published API price or without a reported DeepSWE v1.1 score (see the full comparison). · Verified Oct 20, 2026
Model comparison → Cost calculator → Benchmarks →
What is DeepSWE? · Methodology