How many Claude Code agents can I run in parallel?
By default Claude Code runs up to 20 subagents at once in one session, and agent teams have no hard limit on teammates (Anthropic suggests 3 to 5). What stops you is usage: on Pro and Max every subagent, teammate and parallel session draws on the same 5-hour and weekly limits, on an Anthropic API key they share one organization rate limit, and on Standard Compute each key allows 300 to 2,000 requests a minute on personal plans with no 5-hour or weekly window, so the monthly budget is what you size for.
# One key for the main session and every subagent.
export ANTHROPIC_BASE_URL=https://api.stdcmpt.com
export ANTHROPIC_AUTH_TOKEN=YOUR_API_KEY
export ANTHROPIC_API_KEY=
export ANTHROPIC_MODEL=anthropic/claude-sonnet-5.5
export ANTHROPIC_DEFAULT_OPUS_MODEL=anthropic/claude-opus-5.5
export ANTHROPIC_DEFAULT_SONNET_MODEL=anthropic/claude-sonnet-5.5
export ANTHROPIC_DEFAULT_HAIKU_MODEL=anthropic/claude-haiku-5.5
# Subagents on a cheaper model, at most 8 at once.
export CLAUDE_CODE_SUBAGENT_MODEL=anthropic/claude-haiku-5.5
export CLAUDE_CODE_SUBAGENT_MODEL_FORCE=1
export CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS=8
claude
Set every model variable: Claude Code's built-in names such as claude-sonnet-5-5 return model_not_found on Standard Compute, so a subagent that falls back to an unset one fails. CLAUDE_CODE_SUBAGENT_MODEL_FORCE makes every subagent use that model, even one whose definition names another. A limit of 8 keeps one session from opening 20 large contexts at once; raise it if your tasks split well. For smart routing, set every model variable to standardcompute.
Claude Code's own limits on parallel work
| Setting | Controls | Default |
|---|---|---|
CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS | Subagents running at once in one session | 20. A 21st fails with Concurrent subagent limit reached until one finishes |
CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH | How deep subagents can spawn their own | Three layers below the main conversation; 1 turns nesting off |
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS | Agent teams, off unless set to 1 | No hard limit; 3 to 5 teammates suggested. One team per session, and teammates can't spawn teammates |
Separate sessions, one per git worktree, have no limit from Claude Code: each is a full session. From Anthropic's subagents and agent teams docs, October 14, 2026. The subagent limit applies from Claude Code v2.1.217.
On Pro and Max, every agent shares one limit
Anthropic counts claude.ai, Claude Code and Claude Desktop toward the same limit, and each subagent's requests count toward it like the main conversation's. Five sessions in five worktrees use the 5-hour window about five times as fast as one, and Anthropic puts agent teams at about 7x the tokens of a single session when teammates plan before they code.
| Plan | Price | Shared limits |
|---|---|---|
| Pro | $20/mo | A session limit that resets every 5 hours, plus a weekly limit |
| Max 5x | $100/mo | 5x Pro's per-session allowance, plus a weekly limit |
| Max 20x | $200/mo | 20x Pro's per-session allowance, plus a weekly limit |
Sources: Max plan, usage limits, Claude Code costs, October 14, 2026. At a limit you wait, upgrade, or continue on usage credits at API rates.
On an Anthropic API key, the organization shares one rate limit
- Start tier: 1,000 requests, 2,000,000 uncached input tokens and 400,000 output tokens a minute on Sonnet 5.5, and the same on Opus 5.5, with a $500 monthly spend cap. Build allows 5,000 requests a minute and Scale 10,000.
- Limits apply to the whole organization, so every key, developer and subagent shares them. Workspaces can get lower sub-limits; they can't add capacity.
- Short bursts count: Anthropic says 60 requests a minute may be enforced as one a second, which is how a burst of parallel subagents gets a 429 below the stated limit.
- Cache reads don't count toward input tokens on most models, so long cached sessions fit far more agents than the raw numbers suggest. New organizations may start on a lower Evaluation tier.
On Standard Compute, each key has its own ceiling
Requests are counted per key, per minute and per 10 minutes, and there is no token limit. Over a limit, Claude Code gets HTTP 429 with a Retry-After header and retries after the wait.
| Plan | Per min | Per 10 min | Agents at 10/min |
|---|---|---|---|
| Starter $19 | 300 | 2,500 | 25 |
| Economy $39 | 600 | 5,000 | 50 |
| Standard $89 | 1,000 | 8,000 | 80 |
| Pro $249 | 2,000 | 16,000 | 160 |
Per API key, from the gateway's plan settings on October 14, 2026. Over 10 minutes Starter allows 2,500 requests, so its sustained rate is 250 a minute, and the last column divides that by 10 requests a minute per agent.
What N Claude agents cost for a day
| Agents | Sonnet median | Sonnet heavy | Luna median | Budget for 22 days |
|---|---|---|---|---|
| 1 agent | $3.53 | $28.97 | $0.23 | Standard ($95) |
| 3 agents | $10.59 | $86.91 | $0.69 | Pro ($269) |
| 5 agents | $17.65 | $144.85 | $1.15 | Pro Plus, company ($549) |
| 10 agents | $35.30 | $289.70 | $2.30 | Growth, company ($1,119) |
Day costs for Claude Sonnet 5.5 and GPT-6 Luna at list prices from our gateway's traffic data (Agent Burn Index, September 13 to October 12, 2026): the median and the heavy (90th-percentile) coding-agent day. The index measures one account's day, which this table treats as one agent's day, so read it as an estimate. Last column: the smallest Standard Compute plan whose compute budget covers 22 median Sonnet 5.5 days for every agent if every request goes to Sonnet 5.5; with routing, cheaper models take part of the work.
Set Claude Code up for parallel work
- Pick a plan from $19 a month and create a key, or run
standardcompute launch claude. - Set every model variable, and put subagents on a cheaper model with
CLAUDE_CODE_SUBAGENT_MODELandCLAUDE_CODE_SUBAGENT_MODEL_FORCE=1(tabs above). Subagents mostly read and search, which rarely needs Opus. - Lower
CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTSif one session starts more subagents than you need. Each subagent opens its own context, and that context is what you pay for. - For separate tasks, run one session per git worktree. In a company workspace, give each worktree its own personal key on your plan to see what each task costs.
The catch
Parallel agents spend in parallel: every session and subagent draws on the same budget, and an Opus 5.5 day costs about $7.06 at list prices, twice a Sonnet 5.5 day. A flat plan stops new requests at 100% of its budget, so put subagents on a cheaper model or choose a budget that covers the fleet. On Standard Compute, Claude Code's thinking toggle and Anthropic's server tools such as web search are not passed through.
Where Standard Compute fits
Standard Compute serves the Anthropic Messages API at https://api.stdcmpt.com, so Claude Code, its subagents and agent teams run on it unchanged, with Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1 next to GPT, Gemini, DeepSeek, GLM and Kimi. Plans cost $19 to $249 a month for individuals, each with a fixed compute budget and nothing billed on top. It is a separate API service and does not include a Claude subscription.
FAQ
How many subagents can Claude Code spawn?
20 running at once per session by default; a spawn beyond that fails with Concurrent subagent limit reached and succeeds again once one finishes. Change it with CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS. Subagents can spawn their own, up to three layers below the main conversation (CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH).
Can I run multiple Claude Code instances at once on Max?
Yes. Anthropic documents no cap on sessions, but every instance draws on the same session limit that resets every five hours and the same weekly limit, together with claude.ai and Claude Desktop. Four instances reach the limit roughly four times as fast as one.
Do subagents count toward my Claude usage limit?
Yes. Anthropic's docs say each subagent sends its own requests, which count toward the same usage limits as the main conversation. /usage on Pro and Max shows the share that went to subagents.
Agent teams or subagents: which uses more?
Agent teams. Each teammate is a separate Claude Code instance with its own context window, so token use grows with every teammate. Subagents return only a summary to the main conversation, so they cost less for focused jobs.
Will 10 parallel Claude Code sessions hit Standard Compute's rate limit?
Not at a normal pace. Ten sessions at 10 requests a minute send 100 a minute; the $19 Starter key allows 300 a minute and 2,500 per 10 minutes. The budget is the real limit, as the cost table above shows.
Which model should Claude Code subagents use?
A cheaper one. Set CLAUDE_CODE_SUBAGENT_MODEL to a model such as anthropic/claude-haiku-5.5, and CLAUDE_CODE_SUBAGENT_MODEL_FORCE=1 so subagent definitions can't override it. Keep Opus or Sonnet for the main session, where planning and edits happen.