HASTATAgent stack guide

Agent stack recommender

Choose the stack, not just the model.

Match your harness, workload, billing constraints, context, and budget to a practical setup you can run now.

Recommendation beta

Live provider offers joined to dated editorial evidence; never a universal best-model score.

Read the methodology →

Your constraints

API token amounts are uncached planning scenarios, not spend predictions or subscription quotas. Subscription activity profiles remain separate; cache discounts are never applied.

Recommendation results

Qwen3.8 Max (hosted)

Fresh · fetched 2026-08-17T02:11:57.739Z · Models.dev

Best API stack

Editorial evidence rank 2 · official-only

A hosted everyday candidate; live offer facts decide eligibility and cost.

Published ranking rule: Balanced: ranks through bestRank + 1; then lower uncached estimate, evidence class (independent, mixed, official-only), editorial rank, direct provider, and stable provider/model/evidence IDs.

Primary model
Qwen3.8 Max (hosted)Alibaba (alibaba) · 1M context$2.00 input$6.00 output/ 1M$0.25 cache read· $2.50 cache write / 1M
Planning estimate
$26 / month10M input + 1M output / month · uncached input + output; not a spend prediction
Budget fit
Within monthly budget $100
Background role
DeepSeek V4 FlashAlibaba Cloud (alibaba) · $0.20 input · $0.40 output / 1M · $2.40 full-volume estimate
Vision role
Qwen3.8 Max (hosted)Alibaba Cloud (alibaba); exact offer reports image input
Connection facts
Alibaba Cloud · qwen3.8-maxProvider documentation · Hosted Qwen route with documented OpenAI-compatible harness configuration.

Background uses DeepSeek V4 Flash via Alibaba Cloud; its full-volume estimate is cheaper than the primary. Vision uses Qwen3.8 Max (hosted) via Alibaba Cloud; the exact offer reports image input.

Evidence sources (2)

Caveats

  • Official-only evidence is ranked conservatively in ties.
  • This entry matches hosted Max exact IDs only, not the raw 2.4T checkpoint.

Eligible alternatives

GPT-5.6 SolOpenAI (openai) · 1.05M context$5.00 input · $30.00 output / 1M$80 · Editorial evidence rank 1Higher uncached estimate within the allowed editorial rank window.

Curated subscription option

verified

Codex Pro, one subscription

A curated one-account option for Hermes, Pi, and Codex CLI when subscription routing matters more than API flexibility.

Codex Pro is a curated route for everyday interactive coding.

Actual monthly price
$100/monthCodex Pro 5x from $100/month; the reviewed 1M-context setup and approximate five-hour message range are planning bounds, not guarantees.
Plan type
Included allowance
Model scope
Codex plan-selected models; GPT-5.6 Sol is the reviewed primary route.1M reviewed recommendation context
Key limits
Codex Pro 5x from $100/month; the reviewed 1M-context setup and approximate five-hour message range are planning bounds, not guarantees.

Caveats

  • Weekly limits may apply.
  • Actual usage depends on context, reasoning, tools, retrieval, and caching.
  • Subscription quotas cannot be compared directly with API tokens.

Official subscription source

Subscription credits, quotas, windows, and API token scenarios are separate units. Nominal vendor credits are never compared across providers.

Eligible alternatives

QwenCloud Token Plan Individual Lite$6/month · Quota / windowHosted qwen3.8-max plus the current models listed by the individual plan.2,500 model credits / weekLess-preferred workload rank (2 vs 1).
  • Individual plans are interactive-only, single-device personal use; backend and batch automation are prohibited.
  • Use can involve Singapore/global data transfer.
  • Seven-day quota windows and current prices can change; credits are not API tokens.
  • The reviewed primary model is qwen3.8-max; the official table also lists current DeepSeek and GLM routes.

Official plan source·Official plan source·Official plan source

Live decision data

Inspect the model and provider catalogue.

Compare distinct models without losing provider-specific offers, prices, and caveats.

ModelsProviders