The Token Bill
Live leaderboard of AI API deals — pricing, subscriptions, and SWE-bench scores for Claude, GPT, Gemini, Kimi, DeepSeek, Llama and more. Compare 14 benchmarked models across 16 hosts by price-per-performance, refreshed daily.
| # | Provider | Model | Context | SWE-bench | Input $/MTok | Output $/MTok | Type |
|---|---|---|---|---|---|---|---|
| 1 | AK | — | 1361.7% | $0.12 | $0.58 | Open | |
| 2 | IN | 196k | 1383.9% | $0.15 – $0.47 | $0.90 – $1.86 | Closed | |
| 3 | TE | — | 1324.5% | $0.30 – $0.96 | $0.50 – $2.87 | Open | |
| 4 | — | 1487.6% | $0.30 – $0.40 | $1.20 – $2.00 | Closed | ||
| 5 | 197k | 1398.1% | $0.30 | $1.20 | Closed | ||
| 6 | — | 1387.0% | $0.36 – $0.47 | $1.44 – $1.86 | Closed | ||
| 7 | — | 1297.0% | $0.35 – $0.47 | $1.41 – $1.86 | Closed | ||
| 8 | SC | — | 1249.8% | $0.29 | $1.75 | Open | |
| 9 | TE | — | 1594.8% | $0.40 – $0.80 | $2.40 – $4.00 | Open | |
| 10 | 1.0M | 1449.1% | $0.30 | $2.50 | Closed | ||
| 11 | TE | — | 1435.9% | $0.50 – $0.93 | $2.80 – $4.65 | Open | |
| 12 | 1.0M | 1382.8% | $0.50 | $3.00 | Closed | ||
| 13 | — | 1434.3% | $0.72 – $0.93 | $2.64 – $3.41 | Closed | ||
| 14 | IN | 262k | 1508.7% | $0.73 – $1.00 | $3.50 – $4.00 | Open | |
| 15 | 1.0M | 1527.8% | $0.75 | $3.75 | Closed | ||
| 16 | IN | 262k | 1472.7% | $0.75 – $1.25 | $3.50 – $4.50 | Open | |
| 17 | IN | 1.0M | 1593.3% | $1.20 – $2.10 | $4.20 – $6.42 | Closed | |
| 18 | 203k | 1435.5% | $1.20 – $1.55 | $3.84 – $4.96 | Closed | ||
| 19 | DeepSeek V4 Pro 04235 hosts | 1.0M | 1463.6% | $1.32 – $1.75 | $3.48 – $3.96 | Open | |
| 20 | 200k | 1327.4% | $1.00 | $5.00 | Closed | ||
| 21 | 1.0M | 1506.4% | $1.50 | $9.00 | Closed | ||
| 22 | 1.0M | 1225.7% | $1.25 | $10.00 | Closed | ||
| 23 | Kimi K32 hosts | 1.0M | 1673.7% | $3.00 | $15.00 | Open | |
| 24 | 1M | 1522.5% | $3.00 | $15.00 | Closed | ||
| 25 | 1.1M | 75.7% | $0.20 | $1.20 | Closed |
All subscription deals
Hand-curated catalog of every subscription, flat-rate, and free-harness plan worth knowing about for coding use in 2026. Grouped by how they connect to harnesses.
Flat-rate API subscriptions
API key from a monthly subscription. Best value for power users of external harnesses — no per-token metering.
Flat weekly rate for high-volume Kimi K2.5 Turbo access. API key works in any OpenAI-compatible harness.
- Models
- Kimi K2.5 Turbo
- Harness
- OpenCode, Cline, Roo, any OpenAI-compat
Flat monthly rate giving API access to Claude-class frontier models. Drop the key into any agent framework.
- Models
- Claude-class frontier
- Harness
- Cline, Roo, OpenCode, any framework
Entry tier of MiniMax's flat-rate API bundle. Uses standard API keys so any harness works.
- Models
- MiniMax M2.7
- Harness
- OpenCode Zen, Cline, Roo, OpenClaw
Mid-tier MiniMax subscription with 4,500 requests per 5-hour window. Best price/volume ratio in the flat-rate bucket.
- Models
- MiniMax M2.7 + speech/image
- Harness
- OpenCode Zen, Cline, Roo, OpenClaw
Top MiniMax tier, 15,000 requests per 5-hour window plus all modalities. For continuous agent workflows.
- Models
- MiniMax M2.7 + all modalities
- Harness
- OpenCode Zen, Cline, Roo, OpenClaw
Pay-as-you-go gateway to multiple frontier models at zero markup. No monthly fee — just top up and use.
- Models
- Kimi K2.5, GPT-5-Codex, Gemini, Claude
- Harness
- OpenCode native, any via key
Consumer subscriptions (harness-compatible)
Standard chatbot plans that can be unlocked for harness use via subscription auth plugins. You keep the chat app, you also get the CLI.
Standard Claude chat subscription. Claude Code subscription auth is reusable by OpenCode, Cline, and other harnesses. Some third-party restrictions apply on Max tokens.
- Models
- Claude Sonnet 4.6
- Harness
- Claude Code, OpenCode, Cline (sub auth)
- Own agent
- Claude Code
~225 Sonnet messages per 5-hour window (5× Claude Pro). Claude Code subscription auth reusable by OpenCode, Cline, and other harnesses — though Anthropic restricts some third-party consumers of Max tokens.
- Models
- Claude Sonnet 4.6, Opus 4.6
- Harness
- Claude Code, OpenCode, Cline (sub auth)
- Own agent
- Claude Code
~900 Sonnet messages per 5-hour window (20× Claude Pro). Heaviest Claude Code tier for full-day Opus runs. Subscription auth reusable by OpenCode / Cline with some third-party restrictions.
- Models
- Claude Sonnet 4.6, Opus 4.6
- Harness
- Claude Code, OpenCode, Cline (sub auth)
- Own agent
- Claude Code
Baseline Codex tier. Token-based 5-hour windows (since Apr 2 2026). Codex OAuth works in OpenCode, ForgeCode, Cline and other OpenAI-compatible harnesses.
- Models
- GPT-5, o3, o4-mini, GPT-5 Codex
- Harness
- Codex CLI, OpenCode, ForgeCode, Cline (Codex OAuth)
- Own agent
- Codex CLI
New $100 Pro tier (Apr 9 2026). 5× Codex of Plus — built for heavy agentic coding sessions. Same model access as $200 Pro. 10× boost through May 31 promo.
- Models
- GPT-5, o3, o4-mini, GPT-5 Codex
- Harness
- Codex CLI, OpenCode, ForgeCode, Cline (Codex OAuth)
- Own agent
- Codex CLI
Maximum Codex throughput — 20× Plus limits. Unlimited GPT-5 Instant/Thinking, 250 Deep Research runs/mo. Same harness OAuth as other tiers.
- Models
- GPT-5, o3, o4-mini, GPT-5 Codex
- Harness
- Codex CLI, OpenCode, ForgeCode, Cline (Codex OAuth)
- Own agent
- Codex CLI
Gemini 2.5 Pro consumer tier. Jules and Code Assist ship native; OpenCode authenticates via the gemini-auth plugin for harness use.
- Models
- Gemini 2.5 Pro
- Harness
- Jules, Code Assist, OpenCode (gemini-auth)
- Own agent
- Jules, Code Assist
Formerly ChatGPT Team. Per-seat pricing with admin console, SOC 2, SAML SSO. Standard seats include full Codex access with token-based credits.
- Models
- GPT-5, o3, o4-mini, GPT-5 Codex
- Harness
- Codex CLI, OpenCode, ForgeCode, Cline (Codex OAuth)
- Own agent
- Codex CLI
IDE / terminal native agents
Ship their own harness in-editor. Most also allow BYOK so you can drive external tools with the same plan.
Intent agent for VS Code / JetBrains / CLI. 40k credits/mo, uses Augment's own models — no BYOK.
- Models
- Augment proprietary
- Harness
- Intent CLI (internal)
- Own agent
- Intent
Team tier: 130k credits/mo, up to 20 users. Same locked-in Augment models.
- Models
- Augment proprietary
- Harness
- Intent CLI (internal)
- Own agent
- Intent
Heavy-use tier: 450k credits/mo. Still locked to Augment's models.
- Models
- Augment proprietary
- Harness
- Intent CLI (internal)
- Own agent
- Intent
Terminal AI agent with BYOK unlocked. 1,500 credits/mo.
- Models
- Multi (BYOK)
- Harness
- Warp Agent + BYOK
- Own agent
- Warp Agent
10,000 credits/mo for Warp's terminal agent, with BYOK support.
- Models
- Multi (BYOK)
- Harness
- Warp Agent + BYOK
- Own agent
- Warp Agent
Top Warp tier at 18,000 credits/mo. BYOK available.
- Models
- Multi (BYOK)
- Harness
- Warp Agent + BYOK
- Own agent
- Warp Agent
Cursor IDE with Claude + GPT agent mode. BYOK supported for external API use.
- Models
- Claude + GPT (multi)
- Harness
- Cursor IDE + BYOK
- Own agent
- Cursor Agent
10,000 premium requests per month across all frontier models. Top tier for Cursor heavy users.
- Models
- All frontier (multi)
- Harness
- Cursor IDE + BYOK
- Own agent
- Cursor Agent
Windsurf IDE agent with 500 credits/mo and BYOK support.
- Models
- Multi (BYOK)
- Harness
- Windsurf IDE + BYOK
- Own agent
- Windsurf Agent
Rolling-window power user tier for Windsurf. BYOK available.
- Models
- Multi (BYOK)
- Harness
- Windsurf IDE + BYOK
- Own agent
- Windsurf Agent
Copilot with agent mode in VS Code + JetBrains. 300 advanced requests, MCP-compatible, BYOK supported.
- Models
- Multi (BYOK)
- Harness
- VS Code + JetBrains + MCP + BYOK
- Own agent
- Copilot Agent
Higher Copilot tier with access to all frontier models and higher request limits.
- Models
- All frontier (multi)
- Harness
- VS Code + JetBrains + MCP + BYOK
- Own agent
- Copilot Agent
BYOK harnesses (free platform)
Free (or near-free) harness platforms. Pair these with any API key from the sections above.
Open-source VS Code + CLI agent. Free for individuals, $20/mo for teams. Drop in any provider key.
- Models
- Any (BYOK)
- Harness
- VS Code + CLI, any provider key
- Own agent
- Cline Agent
Free terminal TUI agent. Plug in any provider or use Zen as an optional gateway.
- Models
- Any (BYOK)
- Harness
- Any provider key, Zen optional
- Own agent
- OpenCode Agent
Free VS Code / JetBrains plugin. Works with any OpenAI-compatible provider.
- Models
- Any (OpenAI-compat)
- Harness
- VS Code + JetBrains, any OpenAI-compat key
- Own agent
- Continue Agent