Flat rate. Every model.
No token math.
Every dollar you pay turns into $3 of model usage at provider rates — and roughly $6 when you pair DevPass with SoulForge.
Lite
For occasional AI-assisted coding
- All 200+ models — Claude, GPT-5, Gemini, GLM, Qwen, …
- Works with Claude Code, OpenCode, SoulForge & every OpenAI-compatible tool
- Real-time usage dashboard with per-request cost
- Switch tiers any time — prorated
Pro
For daily development workflows
- Everything in Lite
- Headroom for full-day agent runs in Claude Code & OpenCode
- Priority routing on flagship models
- Email support with 1-business-day reply
Max
For power users and heavy sessions
- Everything in Pro
- Comfortable for non-stop SoulForge & Claude Code usage
- Priority support, faster turnaround
- Best $/usage ratio across the lineup
Usage is metered at each provider's published per-token rate. Every request shows its dollar value in your dashboard in real time.
Cut ~50% of tokens.
Double the value of your DevPass.
SoulForge is a coding agent built around aggressive prompt caching and context reuse. Point it at LLM Gateway and it sends roughly half the tokens of an equivalent Claude Code session — same model, same task, smaller bill.
- 1Prompt caching by default — system prompt, tools, and project context are cached on every provider that supports it.
- 2Context-aware compaction — SoulForge prunes stale turns instead of replaying the whole conversation.
- 3Same DevPass key — no separate subscription. Run
soulforge, type/keys, paste your key.
Standard agent loop, no aggressive caching
Prompt-cache hits on every reusable prefix
Actual savings vary by workload. The 50% figure is typical for multi-turn agent sessions where the system prompt and codebase context stay stable.
Compare plans
What's in each plan
Every tier ships with the full model catalog. The only thing that changes is how much usage you have to play with each month.
| Feature | Lite $29/mo · $290/yr | ProPOPULAR $79/mo · $790/yr | Max $179/mo · $1790/yr |
|---|---|---|---|
| You pay | $29/mo | $79/mo | $179/mo |
| Monthly model usage at provider rates | $87 | $237 | $537 |
| Effective with SoulForge (~50% token cut) | ~$174 | ~$474 | ~$1074 |
| Models included | 200+ | 200+ | 200+ |
| Latest flagships (Opus 4.7, GPT-5.5, Gemini 3.1 Pro) | Included | Included | Included |
| Open-weight Chinese coders (GLM-4.7, Qwen3, Kimi K2.6) | Included | Included | Included |
| Works with Claude Code, OpenCode, SoulForge | Included | Included | Included |
| Any OpenAI/Anthropic-compatible tool | Included | Included | Included |
| Real-time usage dashboard | Included | Included | Included |
| Per-request cost & latency analytics | Included | Included | Included |
| Switch tiers anytime (prorated) | Included | Included | Included |
| Annual billing (save 2 months) | Included | Included | Included |
| Email support | Included | Included | Included |
| Priority support | Not included | Included | Included |
| Headroom for all-day agent runs | Not included | Not included | Included |
Usage is metered at each provider's published per-token rate (input, output, and cached tokens). Every request shows its dollar value in your dashboard in real time. SoulForge savings vary by workload — 50% is typical for multi-turn agent sessions where the system prompt and codebase context stay stable.
Not sure which plan fits?
Estimate your monthly cost with the token calculator — pick a model, paste a sample prompt, and see what your usage actually looks like.
FAQ
Common questions
Everything you need to know about usage limits, plans, and getting started with DevPass.
Can't find an answer? Contact us
Every plan includes a fixed monthly usage allowance, calculated at standard provider rates:
- Lite — $29/mo: $87 in monthly model usage
- Pro — $79/mo: $237 in monthly model usage
- Max — $179/mo: $537 in monthly model usage
That's roughly 3× the subscription price — effectively model access at a deep discount compared to paying providers directly.
Still deciding?
Start on Pro — most developers ship from there. Switch tiers any time, prorated.