↓ Skip to main content
tomrochette.com
  1. Agents/
  2. Model access/

OpenCode Zen

Author
glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 6 minutes read (~1006 words)
agent-curated 544 ai-gateway 3 fully-ai-generated 647 llm=glm-5.3-flash 555 model-access 28 pay-per-use 1 research-note 514
Table of Contents

OpenCode Zen is the OpenCode (Anomaly) team’s curated pay-per-use AI gateway, one API key over a tested catalog that spans open models and the Claude, GPT, Gemini, and Grok lines.

What it is
#

Zen is a billing and routing gateway, not a model: you add credits, get a key, and call OpenAI-compatible, Anthropic-compatible, Google, or Responses endpoints under opencode.ai/zen/v1. The team benchmarks each model/provider pair and serves what passes, their answer to getting a degraded version of a model through a generic router. It also sells to teams (workspaces, roles, member spending caps, model toggles, BYOK for OpenAI and Anthropic keys), and free stealth and promo models rotate through the catalog.

Status
#

Active, with heavy catalog churn managed through a public deprecation table (docs updated 2026-10-10), and workspaces free during the beta with team pricing unannounced. The Reddit thread “Opencode Zen is astoundingly more expensive than OpenRouter” (July 2026) is the sharpest public criticism and remains the note’s key stress test. Zen sells curation and reliability, and community measurements say that premium can exceed 4x the cheapest gateway on identical models.

Strengths
#

  • Every endpoint is benchmarked for coding-agent use, so you avoid badly served providers.
  • A stable of free models (Big Pickle, Space Bunny Free, MiMo-V2.6-Flash Free, Nemotron 3 Ultra Free, among fourteen free rows as of 2026-10-14) with documented data caveats.
  • Markups are claimed to be zero beyond pass-through processing fees (4.4% + $0.30 per card transaction).
  • US- and EU-hosted per the docs as of 2026-10-11, with zero-retention on most models and exceptions listed per model.
  • Same balance backs the Go subscription’s overflow, and the key works from any agent.

Cautions
#

  • A July 2026 community test measured Zen at more than 4x OpenRouter on Kimi K2.6 and MiniMax M2.7.
  • The gap persists per model as of 2026-09-26: Zen lists GLM-5.3-Flash at $0.15/$0.50 per 1M tokens while OpenRouter’s catalog shows about $0.05/$0.14.
  • Auto-reload adds $20 whenever your balance drops below $5, which can overrun a monthly budget you set.
  • Deprecated models vanish on published dates, so configs need migrations, and free stealth models may use your data to improve the model during their free window.

Pricing
#

Pay-as-you-go per 1M tokens, as of 2026-10-11: GLM-5.3 $1.40/$4.40, GLM-5.3-Flash $0.15/$0.50, Kimi K3 $3/$15, Qwen3.7 Plus $0.40/$1.60, DeepSeek V4.1 Flash $0.30/$1.20, MiniMax M3 $0.30/$1.20. Frontier lines: Claude Sonnet 5 $2/$10, Claude Opus 5.5 $4/$20, Claude Fable 5.1 $10/$50, GPT 5.5 $5/$30, GPT 6 Astra $10/$50, GPT 6 Sol $2/$10, GPT 6.1 Sol $2/$10, Gemini 3.8 Flash $1.50/$7.50, Grok 4.7 $2/$6, with long-context surcharge tiers documented per model. Mistral Large 4 currently carries a 50%-off promo row ($0.68/$2.09, struck from $1.36/$4.18), Gemini 4 Argon carries a second one ($2.00/$10.00, struck from $4.00/$20.00, through November 30, 2026), and Jev 1.13 charges $0.042 per input with free output. Fourteen models are free as of 2026-10-14, including the Big Pickle, Space Bunny, and Cat Jam stealth models and a Jev 1.13 Free row, all for a limited time. Auto-reload charges $20 when the balance falls below $5, and OpenCode itself uses low-cost models to generate session titles on your bill.

Price history
#

Date Plan Change Source
2026-09-26 Pay-per-use catalog Baseline: per-model rates as published (GLM-5.3 $1.40/$4.40, Kimi K3 $3/$15, Jev 1.13 $0.042 in) https://opencode.ai/docs/zen/

Compared to
#

OpenCode Go is the sibling subscription: $10/month for capped open-model usage, no frontier models, same balance; choose it when you want a ceiling. OpenRouter is the breadth play: hundreds of models, passthrough pricing, 5.5% credit fee; choose it when price per token on open models matters more than curation. Holding your own provider keys is cheapest for a single provider, and Zen’s BYOK only covers OpenAI and Anthropic.

Bottom line
#

Recommended for OpenCode-centric developers who want vetted endpoints, free models to fall back on, and US-hosted zero-retention. Not for price-sensitive, high-volume runs on open models, where OpenRouter wins on raw per-token cost. My disagreeable claim: the curation premium is worth paying only when a mis-served provider would silently degrade your agent, and most coding work is not that sensitive.

Changes
#

  • 2026-09-26 - Created.
  • 2026-09-29 - Catalog check: the free roster grew from nine to ten models (LongCat 2.5 Preview Free joined), Claude Fable 5.1 and the GPT 6 line appeared on the price table, and the docs republished 2026-09-28; every previously listed per-token rate is unchanged.
  • 2026-10-04 - Free roster moved from ten to twelve models on the live price table (Fledge Alpha Free newly listed as limited-time); every per-token rate re-verified unchanged.
  • 2026-10-06 - Converted the Compared-to cross-references from plain-text paths into working links.
  • 2026-10-07 - Catalog check: the price table’s free rows count thirteen (from twelve recorded 2026-10-04, with Nemotron 3.5 Lightning Free and Muse Spark 1.3 Contributor Free now on the table), Mistral Large 4 gained a 50%-off promo row, and the docs now state US and EU hosting where the note said US-hosted; every per-token rate re-verified unchanged.
  • 2026-10-14 - Catalog check: the free roster moved to fourteen rows with Cat Jam Free joining as a third stealth free model, and Gemini 4 Argon gained a 50%-off promo row (through November 30, 2026) beside Mistral Large 4’s; every previously listed per-token rate re-verified unchanged.

See also
#

  • OpenCode Go - the sibling subscription that overflows into this balance.
  • OpenRouter - the broader, usually cheaper gateway this one is measured against.
  • OpenCode - the harness Zen is the default recommended provider for.
  • Model provider feature matrix - where Zen sits among access providers.

References
#