MINTOK
1658 CODING SCORE
LOG IN BUY PRO ($9.99)
ARENA CODING LEADERBOARD // 1658 FRONTIER SCORE

AUTONOMOUS CODING AGENTS
AT 1/6TH THE INFERENCE COST

MINTOK 1 MAX (1658 CODING SCORE | QWEN-3.8-MAX-0902) AT 0.44¢ PER TASK. 9.3X CHEAPER THAN CLAUDE 5.5 SONNET. 40.7X CHEAPER THAN GPT-6 ASTRA MAX.
DROP-IN OPENAI API COMPATIBLE. FOUR PREDICTABLE MONTHLY PLANS: $9.99 / $19.99 / $39.99 / $99.99.

BUY PRO FOR $9.99 / MO DOWNLOAD DESKTOP & CLI ↓ SEE ARENA RANKINGS →
QUICK INSTALL: curl -fsSL https://mintok.adstim.net/install | bash | npm, brew, desktop →
DROID-ROUTER-01 // PRODUCTION WORKLOAD // 5,073 IN + 1,691 OUT
STANDARDIZED BENCHMARK TASK
mintok [DROID]> /goal refactor distributed lock to raft consensus
ROUTER: High complexity detected -> Dispatched to MinTok 1 Max (1658 Coding Score, qwen-3.8-max-0902 foundation)
COMPILER: Analyzed 4,368 AST nodes -> Sliced to 142 relevant symbols (96.8% irrelevant code eliminated)
VIRTUALIZER: 14,280 lines of test stderr virtualized into 4 minimal stack frames
VERIFIED in 5,073 input + 1,691 output tokens | Cost: $0.00436 on MinTok 1 Max (vs $0.04058 Claude 5.5 Sonnet)
RESULT: 229 VERIFIED TASKS PER $1.00 (9.3X COST ADVANTAGE OVER FRONTIER APIS)
ARENA.AI CODE ARENA // WEBDEV PARETO

THE ARENA CODING PARETO FRONTIER: QUALITY VS. PRICE

RECONSTRUCTED WITH 120+ EVALUATED ARENA CODING MODELS. MINTOK CREATES A NEW UPPER-RIGHT DOMINANT BOUNDARY.

CATEGORY: CODE / WEBDEV // LATEST PARETO

Pareto Frontier

Model performance at each price point on web development tasks
Arena Score 1800 1600 1400 1200 $50 $20 $10 $5 $2 $1 $0.50 $0.20 $0.10 O GPT-6 Astra Max A\ Claude Opus 5.5 A\ Claude 5.5 Sonnet DS DeepSeek V4.1 Flash M MinTok 1 Max 1658 Score · $0.65/M M MinTok 1 Pro 1638 Score · $0.12/M M MinTok 1 Flash 1615 Score · $0.08/M Blended price per 1M tokens (3:1 Ratio)
Pareto Optimal Models
Latest frontier models evaluated on real-world coding benchmarks
MINTOK ACTIVE FRONTIER
M
MinTok 1 Max // #1 Autonomous Coding Frontier (qwen-3.8-max-0902)
MinTok Optimized · 229 verified tasks / $1.00
1658
$0.65/M
M
MinTok 1 Pro // Optimal Agentic Workhorse
MinTok Optimized · 1,194 verified tasks / $1.00
1638
$0.12/M
M
MinTok 1 Flash // Ultra-High Velocity
MinTok Optimized · 1,972 verified tasks / $1.00
1615
$0.08/M
A\
Claude Opus 5.5 (Max)
Anthropic · $16.00 in / $64.00 out · 5 tasks / $1 (43.4× More Expensive)
1818
$28.00/M
O
GPT-6 Astra Max
OpenAI · $15.00 in / $60.00 out · 6 tasks / $1 (40.7× More Expensive)
1789
$40.00/M
A\
Claude Fable 5.1 (Max)
Anthropic · $10.00 in / $40.00 out · 8 tasks / $1 (27.1× More Expensive)
1751
$25.00/M
O
GPT-6.1 Sol (Max)
OpenAI · $4.00 in / $16.00 out · 21 tasks / $1 (10.9× More Expensive)
1759
$8.00/M
A\
Claude 5.5 Sonnet (High)
Anthropic · $3.00 in / $15.00 out · 25 tasks / $1 (9.3× More Expensive)
1709
$6.00/M
G
Gemini 3.1 Pro (Google)
Google · $1.25 in / $5.00 out · 68 tasks / $1 (3.4× More Expensive)
1685
$3.12/M
DS
DeepSeek V4.1 Flash
DeepSeek · $0.35 in / $0.95 out · 296 tasks / $1 · Open Baseline
1620
$0.97/M
THE MINTOK MODEL PORTFOLIO

THREE PRECISION MODELS: FRONTIER QUALITY AT COMMODITY RATES

ACCESSIBLE VIA OPENAI-COMPATIBLE API, MINTOK CLI, AND DIRECT INTEGRATIONS.

FRONTIER ARCHITECTURE 1658 CODING SCORE

MinTok 1 Max

Powered by compiled qwen-3.8-max-0902 foundation for complex distributed systems, multi-file refactors, and mission-critical code.

TRANSPARENT API PRICING:
$0.42 in / $1.32 out / M
~$0.645 / M blended (3:1 ratio)
• Foundation Target: qwen-3.8-max-0902
• Cost per Coding Task: $0.00436 (~0.44¢)
• Advantage: 9.3× cheaper than Claude 5.5 Sonnet
• Advantage: 43.4× cheaper than Claude Opus 5.5
• Context Window: 131,072 tokens
MOST POPULAR // SWE WORKHORSE
DAILY PRODUCTION CODING 1638 CODING SCORE

MinTok 1 Pro

For daily software development, feature delivery, bug squashing, and SWE benchmarks.

TRANSPARENT API PRICING:
$0.085 in / $0.24 out / M
~$0.124 / M blended (3:1 ratio)
• Cost per Coding Task: $0.000837 (~0.084¢)
• Advantage: 48.5× cheaper than Claude 5.5 Sonnet
• Advantage: 56.6× cheaper than GPT-6.1 Sol
• Context Window: 65,536 tokens
HIGH-SPEED ITERATION 1615 CODING SCORE

MinTok 1 Flash

For instant unit test generation, scaffolding, linter autofix, and quick code edits.

TRANSPARENT API PRICING:
$0.05 in / $0.15 out / M
~$0.075 / M blended (3:1 ratio)
• Cost per Coding Task: $0.000507 (~0.051¢)
• Advantage: 1,972 verified tasks per $1.00
• Advantage: 29.1× cheaper than Gemini 3.1 Pro
• Context Window: 32,768 tokens
EXACT BENCHMARK TRAJECTORY ECONOMICS

STANDARDIZED CODING-TASK COST COMPARISON

EVALUATING EXACT INFERENCE COSTS FOR A STANDARDIZED 5,073 INPUT + 1,691 OUTPUT TOKEN CODING TASK.

LATEST FRONTIER MODEL INPUT / M OUTPUT / M TASK COST TASKS / $1.00 MINTOK 1 MAX DELTA
Claude Opus 5.5 (Max) $16.00 $64.00 $0.18939 5 tasks 43.4× More Expensive
GPT-6 Astra Max $15.00 $60.00 $0.17756 6 tasks 40.7× More Expensive
Claude Fable 5.1 (Max) $10.00 $40.00 $0.11837 8 tasks 27.1× More Expensive
GPT-6.1 Sol (Max) $4.00 $16.00 $0.04735 21 tasks 10.9× More Expensive
Claude 5.5 Sonnet (High) $3.00 $15.00 $0.04058 25 tasks 9.3× More Expensive
Gemini 3.1 Pro (Google) $1.25 $5.00 $0.01478 68 tasks 3.4× More Expensive
DeepSeek V4.1 Flash $0.35 $0.95 $0.00338 296 tasks Baseline
MinTok 1 Max (1658 Score | Frontier | qwen-3.8-max-0902) $0.42 $1.32 $0.00436 229 tasks REFERENCE BASELINE
MinTok 1 Pro (1638 Score | Workhorse) $0.085 $0.24 $0.00084 1,194 tasks 5.2× Cheaper Than Max
MinTok 1 Flash (1615 Score | Agile) $0.05 $0.15 $0.00051 1,972 tasks 8.6× Cheaper Than Max
CRYPTOGRAPHIC EMPIRICAL AUDIT

AUDITED EMPIRICAL RIGOR: SWE-HOLDOUT-150 PAIRED EVALUATION

150 CANONICAL REPOSITORIES: 143/150 SOLVED FOR $5.43 TOTAL COMPUTE SPEND VS $27.00+ CONTROL BASELINES.

PAIRED 2x2 CONTINGENCY MATRIX N = 150 BENCHMARKS
OUTCOME CONTROL PASS CONTROL FAIL TOTAL
MINTOK PASS 102 (Both Solve) 41 (MinTok Unique) 143 (95.3%)
MINTOK FAIL 2 (Control Unique) 5 (Both Fail) 7 (4.7%)
TOTAL 104 (69.3%) 46 (30.7%) 150
STATISTICAL POWER: McNemar's Test χ² = 33.72 (df=1, p = 2.78 × 10⁻⁹). Net discordant advantage: +39 tasks.
BENCHMARK RECAP

REPRODUCIBLE RUN METRICS

TOTAL RUN COST
$5.43
vs $27.00+ standard
RESOLVE RATE
95.3%
143 / 150 tasks solved
TOKEN REDUCTION
84.7%
7.06x info density
MEAN TURNS
3.4
vs 8.7 standard loop
COMMERCIAL SUBSCRIPTION PLANS

PREDICTABLE CAPACITY: NO OVERAGES, NO TOKEN HOURLY METERS

EACH PLAN INCLUDES MORE CLOUD USAGE CREDITS THAN ITS MONTHLY COST.

INDIVIDUAL ENGINES

Pro

$9.99 / month
Includes $10.49 cloud usage
• 4 concurrent coding agents
• ~12,500 MinTok 1 Pro tasks / mo
• Background tasks & Git worktrees
• MinTok 1 Max, MinTok 1 Pro, & MinTok 1 Flash access
• Monthly usage rollover (1 mo)
• Zero daily token rate limits
PARALLEL AGENTS

Power

$19.99 / month
Includes $20.99 cloud usage
• 8 concurrent coding agents
• ~25,000 MinTok 1 Pro tasks / mo
• Parallel agent execution
• Priority GPU inference scheduling
• Larger task token budgets
• Monthly usage rollover
ENGINEERING FLEET

Max

$39.99 / month
Includes $41.99 cloud usage
• 16 concurrent coding agents
• ~50,000 MinTok 1 Pro tasks / mo
• Dedicated warm GPU pools
• Durable Objects team ledger
• GitHub Org sync & priority queue
• Uncapped task runtime windows
HIGH-SCALE FLEET
AUTONOMOUS CLOUD

Ultra

$99.99 / month
Includes $109.99 cloud usage
• 40 concurrent coding agents
• ~131,000 MinTok 1 Pro / 25,200 MinTok 1 Max tasks
• Dedicated warm GPU inference pool
• Multi-repo AST caching & index
• Webhook streaming & routing SLA
• 24/7 dedicated fleet support
DEVELOPER HARNESS & API

DROP-IN COMPATIBILITY: CLI, OPENAI API, CURSOR, AND CLINE

SUB-150MS TIME TO FIRST TOKEN
# 1. Point Cursor, Cline, or Aider to MinTok
export OPENAI_BASE_URL="https://mintok.adstim.net/v1"
export OPENAI_API_KEY="mtk_live_..."
# 2. Or execute via MinTok CLI
$ mintok run "Refactor payment webhook" --model mintok-1-pro
# 3. Stream completions via standard OpenAI clients
client.chat.completions.create(model="mintok-1-max", stream=True)
PRODUCTION MODEL ENDPOINTS:
mintok-1-max (MinTok 1 Max — 1658 Score) $0.42 / $1.32
mintok-1-pro (MinTok 1 Pro — 1638 Score) $0.085 / $0.24
mintok-1-flash (MinTok 1 Flash — 1615 Score) $0.05 / $0.15
1. CURL INSTALLER
curl -fsSL https://mintok.adstim.net/install | bash
2. NPM PACKAGE
npm install -g mintok
3. HOMEBREW TAP
brew install mintok/tap/mintok
4. DESKTOP APP
macOS, Win, Linux →
MINTOK ENGINE // BRING YOUR OWN KEY (BYOK)

SLASH YOUR MODEL BILLS BY 60%–80% ACROSS EVERY MAJOR PROVIDER

DROP-IN OPTIMIZATION PROXY FOR OPENAI, ANTHROPIC, GEMINI, AND DEEPSEEK. ZERO ARCHITECTURE CHANGES.

01 // DROP-IN OPTIMIZATION PROXY

Use Your Existing API Keys

Point your agents, Cursor, Cline, or Aider to https://mintok.adstim.net/v1. MinTok compresses prompt payloads in-flight before dispatching to OpenAI, Anthropic, Google, or DeepSeek. You maintain direct provider accounts while cutting token burn.

RAW AGENT PROMPT PAYLOAD: 48,900 TOKENS
OPTIMIZED PAYLOAD TO UPSTREAM API: 11,400 TOKENS
76.7% TOKEN CONSUMPTION ELIMINATED · 3.5× FASTER TTFT
02 // UPSTREAM API PASSTHROUGH

Enterprise Data Retention Standard

MinTok performs in-flight compaction strictly inside isolated memory. We never store, log, or persist user source code or prompts to disk. All requests adhere directly to original upstream provider policies (OpenAI, Anthropic, Google) with 100% zero training on customer data.

PERSISTENT CODE DISK LOGGING: ZERO (IN-FLIGHT MEMORY ONLY)
TRAINING ON CUSTOMER DATA: STRICTLY ZERO ACROSS ALL MODELS
ENTERPRISE COMPLIANCE & DIRECT PROVIDER GOVERNANCE
EXPLORE MINTOK ENGINE & BYOK INTEGRATION →
OBSERVABILITY & FLEET MONITORING

DETERMINISTIC REPLAY: DEBUG AGENTS AT THE TRAJECTORY LEVEL

AUDIT EVERY HYPOTHESIS, OPTIMIZATION STAGE, AND TOKEN SAVING WITH FULL TELEMETRY.

ACTIVE AGENTS: 14 / 40 (ULTRA TIER) | AVOIDED TOKENS TODAY: 48,290,140 | ESTIMATED SAVINGS: $1,280.40
FLEET HEALTHY // 100% ONLINE
RUN-9821 // DIFF VIEWER: UNCOMPILED RAW VS MINTOK OPTIMIZED
WITHOUT MINTOK (RAW CONTEXT):
• 18 tool calls
• 42,381 input tokens
• 3 failed hypotheses
• Task Cost: $0.04058
WITH MINTOK (AST COMPILED):
• 4 tool calls
• 5,073 input tokens
• 1 failed hypothesis
• Task Cost: $0.00084