Fully autonomous AI Agents system capable of performing complex penetration testing tasks
Go ai-agentsai-security-toolanthropicautonomous-agentsgolanggptgraphqlmulti-agent-systemoffensive-securityopen-sourceopenaipenetration-testingpenetration-testing-toolsreactsecurity-automationsecurity-testingsecurity-toolsself-hosted
fix(providers): correct model prices from a full official-source audit
Sergey Kozyrenko4ecf2bbac1b9
diff --git a/backend/pkg/providers/bedrock/models.yml b/backend/pkg/providers/bedrock/models.ymlindex 2cabf128..cad8141c 100644--- a/backend/pkg/providers/bedrock/models.yml+++ b/backend/pkg/providers/bedrock/models.yml@@ -220,8 +220,8 @@thinking: falserelease_date: 2025-12-02price:- input: 4.0- output: 12.0+ input: 0.5+ output: 1.5# Moonshot Kimi Series - Multimodal and thinking agent models- name: moonshotai.kimi-k2.5diff --git a/backend/pkg/providers/deepseek/config.yml b/backend/pkg/providers/deepseek/config.ymlindex 6fe8caa2..3d5ce1b3 100644--- a/backend/pkg/providers/deepseek/config.yml+++ b/backend/pkg/providers/deepseek/config.yml@@ -52,9 +52,9 @@ primary_agent:thinking:type: enabledprice:- input: 1.74- output: 3.48- cache_read: 0.0145+ input: 0.435+ output: 0.87+ cache_read: 0.003625assistant:model: deepseek-v4-pro@@ -66,9 +66,9 @@ assistant:thinking:type: enabledprice:- input: 1.74- output: 3.48- cache_read: 0.0145+ input: 0.435+ output: 0.87+ cache_read: 0.003625generator:model: deepseek-v4-pro@@ -80,9 +80,9 @@ generator:thinking:type: enabledprice:- input: 1.74- output: 3.48- cache_read: 0.0145+ input: 0.435+ output: 0.87+ cache_read: 0.003625refiner:model: deepseek-v4-pro@@ -94,9 +94,9 @@ refiner:thinking:type: enabledprice:- input: 1.74- output: 3.48- cache_read: 0.0145+ input: 0.435+ output: 0.87+ cache_read: 0.003625adviser:model: deepseek-v4-pro@@ -108,9 +108,9 @@ adviser:thinking:type: enabledprice:- input: 1.74- output: 3.48- cache_read: 0.0145+ input: 0.435+ output: 0.87+ cache_read: 0.003625reflector:model: deepseek-v4-flash@@ -164,9 +164,9 @@ coder:thinking:type: enabledprice:- input: 1.74- output: 3.48- cache_read: 0.0145+ input: 0.435+ output: 0.87+ cache_read: 0.003625installer:model: deepseek-v4-flash@@ -192,6 +192,6 @@ pentester:thinking:type: enabledprice:- input: 1.74- output: 3.48- cache_read: 0.0145+ input: 0.435+ output: 0.87+ cache_read: 0.003625diff --git a/backend/pkg/providers/deepseek/models.yml b/backend/pkg/providers/deepseek/models.ymlindex 2b9acf48..a4d5cff1 100644--- a/backend/pkg/providers/deepseek/models.yml+++ b/backend/pkg/providers/deepseek/models.yml@@ -10,6 +10,6 @@description: DeepSeek V4 Pro - Higher-tier reasoning model with hybrid thinking/non-thinking modes (default thinking, switchable via extra_body). Suitable for complex logic, mathematical reasoning, and security analysis. Supports JSON output, tool calls, chat prefix completion (beta), and FIM completion (non-thinking only). 1M context, up to 384K output tokens.thinking: trueprice:- input: 1.74- output: 3.48- cache_read: 0.0145+ input: 0.435+ output: 0.87+ cache_read: 0.003625diff --git a/backend/pkg/providers/minimax/config.yml b/backend/pkg/providers/minimax/config.ymlindex c9ddde53..0bdcb964 100644--- a/backend/pkg/providers/minimax/config.yml+++ b/backend/pkg/providers/minimax/config.yml@@ -5,8 +5,9 @@ simple:n: 1max_tokens: 8192price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06simple_json:model: MiniMax-M3@@ -16,8 +17,9 @@ simple_json:max_tokens: 4096json: trueprice:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06primary_agent:model: MiniMax-M3@@ -26,8 +28,9 @@ primary_agent:n: 1max_tokens: 16384price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06assistant:model: MiniMax-M3@@ -36,8 +39,9 @@ assistant:n: 1max_tokens: 16384price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06generator:model: MiniMax-M3@@ -46,8 +50,9 @@ generator:n: 1max_tokens: 32768price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06refiner:model: MiniMax-M3@@ -56,8 +61,9 @@ refiner:n: 1max_tokens: 20480price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06adviser:model: MiniMax-M3@@ -66,8 +72,9 @@ adviser:n: 1max_tokens: 8192price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06reflector:model: MiniMax-M3@@ -76,8 +83,9 @@ reflector:n: 1max_tokens: 4096price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06searcher:model: MiniMax-M3@@ -86,8 +94,9 @@ searcher:n: 1max_tokens: 4096price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06enricher:model: MiniMax-M3@@ -96,8 +105,9 @@ enricher:n: 1max_tokens: 4096price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06coder:model: MiniMax-M3@@ -106,8 +116,9 @@ coder:n: 1max_tokens: 20480price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06installer:model: MiniMax-M3@@ -116,8 +127,9 @@ installer:n: 1max_tokens: 16384price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06pentester:model: MiniMax-M3@@ -126,5 +138,6 @@ pentester:n: 1max_tokens: 16384price:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06diff --git a/backend/pkg/providers/minimax/models.yml b/backend/pkg/providers/minimax/models.ymlindex c01dba6c..9be49760 100644--- a/backend/pkg/providers/minimax/models.yml+++ b/backend/pkg/providers/minimax/models.yml@@ -2,19 +2,22 @@description: MiniMax-M3 - Latest flagship M-series model with ~1M token context for agentic reasoning, tool use, code generation, and long-context tasks.thinking: falseprice:- input: 0.60- output: 2.40+ input: 0.30+ output: 1.20+ cache_read: 0.06- name: MiniMax-M2.7description: MiniMax-M2.7 - Previous flagship model with enhanced reasoning and coding capabilities. Supports tool calling, JSON output, and streaming.thinking: falseprice:- input: 0.40- output: 1.10+ input: 0.30+ output: 1.20+ cache_read: 0.06- name: MiniMax-M2.7-highspeeddescription: MiniMax-M2.7-highspeed - High-speed version of M2.7 for low-latency scenarios. Same capabilities as M2.7 with improved response latency.thinking: falseprice:- input: 0.40- output: 1.10+ input: 0.60+ output: 2.40+ cache_read: 0.06diff --git a/backend/pkg/providers/qwen/models.yml b/backend/pkg/providers/qwen/models.ymlindex cd407844..595c3241 100644--- a/backend/pkg/providers/qwen/models.yml+++ b/backend/pkg/providers/qwen/models.yml@@ -50,8 +50,9 @@thinking: truerelease_date: 2026-04-22price:- input: 0.248- output: 1.485+ input: 0.375+ output: 2.25+ cache_read: 0.075- name: qwen3.6-27bdescription: Qwen3.6 27B Open-Source - Native vision-language model on hybrid architecture, significantly improved performance over 3.5 series at the same scale. Ideal for on-premises pentesting with privacy-sensitive multimodal workflows@@ -60,6 +61,7 @@price:input: 0.6output: 3.6+ cache_read: 0.12# Qwen3.5 Series - Native vision-language with hybrid linear attention + sparse MoE- name: qwen3.5-plus@@ -87,6 +89,7 @@price:input: 0.6output: 3.6+ cache_read: 0.12- name: qwen3.5-122b-a10bdescription: Qwen3.5 122B Open-Source - Large open-source MoE model (122B params, ~10B active) with strong performance balance for complex on-premises security analysis@@ -95,6 +98,7 @@price:input: 0.4output: 3.2+ cache_read: 0.08- name: qwen3.5-35b-a3bdescription: Qwen3.5 35B Open-Source - Efficient open-source MoE model (35B params, ~3B active) optimized for cost-effective inference in continuous security monitoring@@ -103,6 +107,7 @@price:input: 0.25output: 2.0+ cache_read: 0.05- name: qwen3.5-27bdescription: Qwen3.5 27B Open-Source - Native vision-language model with hybrid linear attention and sparse MoE architecture for higher inference efficiency. Ideal for on-premises security analysis with multimodal capabilities@@ -111,6 +116,7 @@price:input: 0.3output: 2.4+ cache_read: 0.06# Qwen3 Series - Flagship hybrid thinking models- name: qwen3-max@@ -166,6 +172,7 @@price:input: 0.3output: 1.5+ cache_read: 0.06- name: qwen3-coder-30b-a3b-instructdescription: Qwen3 Coder 30B Open-Source - MoE code model (30B params, ~3B active) optimized for efficient repository-scale code understanding. Good for cost-effective on-premises security code review and exploit generation@@ -173,6 +180,7 @@price:input: 0.45output: 2.25+ cache_read: 0.09- name: qwen3-coder-480b-a35b-instructdescription: Qwen3 Coder 480B Open-Source - Largest open-source coder MoE (480B params, ~35B active) with maximum coding capability for advanced exploit development, complex vulnerability discovery, and large-scale code refactoring in security research@@ -180,6 +188,7 @@price:input: 1.5output: 7.5+ cache_read: 0.3# Qwen3-VL Series - Vision-language models for browser screenshot analysis and visual reconnaissance- name: qwen3-vl-plus@@ -208,6 +217,7 @@price:input: 1.2output: 4.8+ cache_read: 0.24# Qwen3 Open Source Series - Hybrid thinking dense and MoE models- name: qwen3-next-80b-a3b-thinking@@ -216,6 +226,7 @@price:input: 0.15output: 1.2+ cache_read: 0.03- name: qwen3-next-80b-a3b-instructdescription: Qwen3-Next 80B Instruct Open-Source - Non-thinking variant for efficient instruction-following, ideal for high-throughput security automation tasks@@ -223,6 +234,7 @@price:input: 0.15output: 1.2+ cache_read: 0.03- name: qwen3-235b-a22bdescription: Qwen3 235B Open-Source - Dual-mode MoE model (235B params, ~22B active) supporting both thinking and non-thinking modes for versatile security workloads (thinking-mode output pricing shown)@@ -230,6 +242,7 @@price:input: 0.7output: 8.4+ cache_read: 0.14- name: qwen3-32bdescription: Qwen3 32B Open-Source - Versatile dense model with dual-mode capabilities, suitable for balanced on-premises pentesting workflows@@ -237,6 +250,7 @@price:input: 0.16output: 0.64+ cache_read: 0.032- name: qwen3-30b-a3bdescription: Qwen3 30B Open-Source - Efficient MoE model (30B params, ~3B active) for cost-effective continuous security monitoring (thinking-mode output pricing shown)@@ -244,6 +258,7 @@price:input: 0.2output: 2.4+ cache_read: 0.04- name: qwen3-14bdescription: Qwen3 14B Open-Source - Medium-sized dense model with good performance-cost balance for routine security assessments (thinking-mode output pricing shown)@@ -251,6 +266,7 @@price:input: 0.35output: 4.2+ cache_read: 0.07- name: qwen3-8bdescription: Qwen3 8B Open-Source - Compact dense model optimized for efficient inference in lightweight security tasks (thinking-mode output pricing shown)@@ -258,6 +274,7 @@price:input: 0.18output: 2.1+ cache_read: 0.036- name: qwen3-4bdescription: Qwen3 4B Open-Source - Lightweight model for simple security tasks and high-frequency scanning (thinking-mode output pricing shown)