Skip to content

Fully autonomous AI Agents system capable of performing complex penetration testing tasks

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.yml
index 2cabf128..cad8141c 100644
--- a/backend/pkg/providers/bedrock/models.yml
+++ b/backend/pkg/providers/bedrock/models.yml
@@ -220,8 +220,8 @@
thinking: false
release_date: 2025-12-02
price:
- input: 4.0
- output: 12.0
+ input: 0.5
+ output: 1.5
# Moonshot Kimi Series - Multimodal and thinking agent models
- name: moonshotai.kimi-k2.5
diff --git a/backend/pkg/providers/deepseek/config.yml b/backend/pkg/providers/deepseek/config.yml
index 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: enabled
price:
- input: 1.74
- output: 3.48
- cache_read: 0.0145
+ input: 0.435
+ output: 0.87
+ cache_read: 0.003625
assistant:
model: deepseek-v4-pro
@@ -66,9 +66,9 @@ assistant:
thinking:
type: enabled
price:
- input: 1.74
- output: 3.48
- cache_read: 0.0145
+ input: 0.435
+ output: 0.87
+ cache_read: 0.003625
generator:
model: deepseek-v4-pro
@@ -80,9 +80,9 @@ generator:
thinking:
type: enabled
price:
- input: 1.74
- output: 3.48
- cache_read: 0.0145
+ input: 0.435
+ output: 0.87
+ cache_read: 0.003625
refiner:
model: deepseek-v4-pro
@@ -94,9 +94,9 @@ refiner:
thinking:
type: enabled
price:
- input: 1.74
- output: 3.48
- cache_read: 0.0145
+ input: 0.435
+ output: 0.87
+ cache_read: 0.003625
adviser:
model: deepseek-v4-pro
@@ -108,9 +108,9 @@ adviser:
thinking:
type: enabled
price:
- input: 1.74
- output: 3.48
- cache_read: 0.0145
+ input: 0.435
+ output: 0.87
+ cache_read: 0.003625
reflector:
model: deepseek-v4-flash
@@ -164,9 +164,9 @@ coder:
thinking:
type: enabled
price:
- input: 1.74
- output: 3.48
- cache_read: 0.0145
+ input: 0.435
+ output: 0.87
+ cache_read: 0.003625
installer:
model: deepseek-v4-flash
@@ -192,6 +192,6 @@ pentester:
thinking:
type: enabled
price:
- input: 1.74
- output: 3.48
- cache_read: 0.0145
+ input: 0.435
+ output: 0.87
+ cache_read: 0.003625
diff --git a/backend/pkg/providers/deepseek/models.yml b/backend/pkg/providers/deepseek/models.yml
index 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: true
price:
- input: 1.74
- output: 3.48
- cache_read: 0.0145
+ input: 0.435
+ output: 0.87
+ cache_read: 0.003625
diff --git a/backend/pkg/providers/minimax/config.yml b/backend/pkg/providers/minimax/config.yml
index c9ddde53..0bdcb964 100644
--- a/backend/pkg/providers/minimax/config.yml
+++ b/backend/pkg/providers/minimax/config.yml
@@ -5,8 +5,9 @@ simple:
n: 1
max_tokens: 8192
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
simple_json:
model: MiniMax-M3
@@ -16,8 +17,9 @@ simple_json:
max_tokens: 4096
json: true
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
primary_agent:
model: MiniMax-M3
@@ -26,8 +28,9 @@ primary_agent:
n: 1
max_tokens: 16384
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
assistant:
model: MiniMax-M3
@@ -36,8 +39,9 @@ assistant:
n: 1
max_tokens: 16384
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
generator:
model: MiniMax-M3
@@ -46,8 +50,9 @@ generator:
n: 1
max_tokens: 32768
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
refiner:
model: MiniMax-M3
@@ -56,8 +61,9 @@ refiner:
n: 1
max_tokens: 20480
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
adviser:
model: MiniMax-M3
@@ -66,8 +72,9 @@ adviser:
n: 1
max_tokens: 8192
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
reflector:
model: MiniMax-M3
@@ -76,8 +83,9 @@ reflector:
n: 1
max_tokens: 4096
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
searcher:
model: MiniMax-M3
@@ -86,8 +94,9 @@ searcher:
n: 1
max_tokens: 4096
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
enricher:
model: MiniMax-M3
@@ -96,8 +105,9 @@ enricher:
n: 1
max_tokens: 4096
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
coder:
model: MiniMax-M3
@@ -106,8 +116,9 @@ coder:
n: 1
max_tokens: 20480
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
installer:
model: MiniMax-M3
@@ -116,8 +127,9 @@ installer:
n: 1
max_tokens: 16384
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
pentester:
model: MiniMax-M3
@@ -126,5 +138,6 @@ pentester:
n: 1
max_tokens: 16384
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
diff --git a/backend/pkg/providers/minimax/models.yml b/backend/pkg/providers/minimax/models.yml
index 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: false
price:
- input: 0.60
- output: 2.40
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
- name: MiniMax-M2.7
description: MiniMax-M2.7 - Previous flagship model with enhanced reasoning and coding capabilities. Supports tool calling, JSON output, and streaming.
thinking: false
price:
- input: 0.40
- output: 1.10
+ input: 0.30
+ output: 1.20
+ cache_read: 0.06
- name: MiniMax-M2.7-highspeed
description: MiniMax-M2.7-highspeed - High-speed version of M2.7 for low-latency scenarios. Same capabilities as M2.7 with improved response latency.
thinking: false
price:
- input: 0.40
- output: 1.10
+ input: 0.60
+ output: 2.40
+ cache_read: 0.06
diff --git a/backend/pkg/providers/qwen/models.yml b/backend/pkg/providers/qwen/models.yml
index cd407844..595c3241 100644
--- a/backend/pkg/providers/qwen/models.yml
+++ b/backend/pkg/providers/qwen/models.yml
@@ -50,8 +50,9 @@
thinking: true
release_date: 2026-04-22
price:
- input: 0.248
- output: 1.485
+ input: 0.375
+ output: 2.25
+ cache_read: 0.075
- name: qwen3.6-27b
description: 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: 0.6
output: 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: 0.6
output: 3.6
+ cache_read: 0.12
- name: qwen3.5-122b-a10b
description: 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: 0.4
output: 3.2
+ cache_read: 0.08
- name: qwen3.5-35b-a3b
description: 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: 0.25
output: 2.0
+ cache_read: 0.05
- name: qwen3.5-27b
description: 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: 0.3
output: 2.4
+ cache_read: 0.06
# Qwen3 Series - Flagship hybrid thinking models
- name: qwen3-max
@@ -166,6 +172,7 @@
price:
input: 0.3
output: 1.5
+ cache_read: 0.06
- name: qwen3-coder-30b-a3b-instruct
description: 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: 0.45
output: 2.25
+ cache_read: 0.09
- name: qwen3-coder-480b-a35b-instruct
description: 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: 1.5
output: 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: 1.2
output: 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: 0.15
output: 1.2
+ cache_read: 0.03
- name: qwen3-next-80b-a3b-instruct
description: 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: 0.15
output: 1.2
+ cache_read: 0.03
- name: qwen3-235b-a22b
description: 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: 0.7
output: 8.4
+ cache_read: 0.14
- name: qwen3-32b
description: Qwen3 32B Open-Source - Versatile dense model with dual-mode capabilities, suitable for balanced on-premises pentesting workflows
@@ -237,6 +250,7 @@
price:
input: 0.16
output: 0.64
+ cache_read: 0.032
- name: qwen3-30b-a3b
description: 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: 0.2
output: 2.4
+ cache_read: 0.04
- name: qwen3-14b
description: 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: 0.35
output: 4.2
+ cache_read: 0.07
- name: qwen3-8b
description: 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: 0.18
output: 2.1
+ cache_read: 0.036
- name: qwen3-4b
description: Qwen3 4B Open-Source - Lightweight model for simple security tasks and high-frequency scanning (thinking-mode output pricing shown)