Models Reference
API pricing, context windows, and SWE-Bench scores for coding AI models.
Compiled June 2026.
Anthropic (Claude)
Current as of June 2026. Source: platform.claude.com
| Model | Input /1M | Output /1M | Batch (50% off) | Cache Writes (5m) | 1h Cache Writes | Cache Hits |
|---|---|---|---|---|---|---|
| Fable 5 | $10.00 | $50.00 | ✓ | $12.50/MTok | $20.00/MTok | $1.00/MTok |
| Mythos 5 | $10.00 | $50.00 | ✓ | $12.50/MTok | $20.00/MTok | $1.00/MTok |
| Opus 4.8 | $5.00 | $25.00 | ✓ | $6.25/MTok | $10.00/MTok | $0.50/MTok |
| Opus 4.7 | $5.00 | $25.00 | ✓ | $6.25/MTok | $10.00/MTok | $0.50/MTok |
| Opus 4.6 | $5.00 | $25.00 | ✓ | $6.25/MTok | $10.00/MTok | $0.50/MTok |
| Opus 4.5 | $5.00 | $25.00 | ✓ | $6.25/MTok | $10.00/MTok | $0.50/MTok |
| Sonnet 4.6 | $3.00 | $15.00 | ✓ | $3.75/MTok | $6.00/MTok | $0.30/MTok |
| Haiku 4.5 | $1.00 | $5.00 | ✓ | $1.25/MTok | $2.00/MTok | $0.10/MTok |
| Mythos Preview | $25.00 | $125.00 | — | — | — | — |
Fable 5 (Jun 2026): Next-gen Anthropic flagship, $10/$50 per MTok. Limited availability. Mythos 5 (Jun 2026): Premium reasoning tier at $10/$50 per MTok. Also limited availability. Opus 4.8 (May 28 2026): 88.6% SWE-Bench Verified (#2). Opus 4.7: 87.6%, 64.3% SWE-Bench Pro (#1). Opus 4.5: 80.9%. Opus 4.6: 80.8%. Sonnet 4.6: 79.6%. Mythos Preview: Research-only preview at $25/$125 per MTok. 93.9% SWE-Bench Verified (#1) — very expensive, limited API access.
Google Gemini
Current as of June 2026. Source: ai.google.dev, SWE-Bench
Current Models
| Model | Input /1M | Output /1M | Cached Input | Context | Max Output | SWE-Bench Verified | Notes |
|---|---|---|---|---|---|---|---|
| Gemini 3.1 Pro Preview | $2.00 ($4.00 >200K) | $12.00 ($18.00 >200K) | $0.20 ($0.40 >200K) | 2M | 16K | 80.6% | Preview. Top-tier reasoning. 2M ctx |
| Gemini 3.5 Flash | $1.50 | $9.00 | $0.15 | 1M | 64K | — | GA May 19 2026. Fast, high-volume agentic tasks. 76.2% Terminal-Bench 2.1 |
| Gemini 3 Flash Preview | $0.50 | $3.00 | $0.05 | 1M | 64K | — | Efficient, general-purpose |
| Gemini 3.1 Flash-Lite Preview | $0.25 | $1.50 | $0.025 | 1M | 64K | — | Fast, high-volume agentic tasks |
| Gemini 2.5 Pro | $1.25 ($2.50 >200K) | $10.00 ($15.00 >200K) | — | 2M | 64K | — | Complex reasoning, coding, long docs |
| Gemini 2.5 Flash | $0.30 | $2.50 | — | 1M | 64K | — | Balanced cost and capability |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | — | 1M | 64K | — | Lowest-cost current Gemini route |
Batch / Flex Pricing (50% off)
| Model | Batch Input /1M | Batch Output /1M |
|---|---|---|
| Gemini 3.1 Pro (≤200K) | $1.00 | $6.00 |
| Gemini 3.5 Flash | $0.75 | $4.50 |
| Gemini 3 Flash | $0.25 | $1.50 |
| Gemini 3.1 Flash-Lite | $0.125 | $0.75 |
| Gemini 2.5 Pro (≤200K) | $0.625 | $5.00 |
| Gemini 2.5 Flash | $0.15 | $1.25 |
| Gemini 2.5 Flash-Lite | $0.05 | $0.20 |
Deprecated
| Model | Input /1M | Output /1M | Note |
|---|---|---|---|
| Gemini 2.0 Flash | $0.10 | $0.40 | Retired June 1 2026 — shutdown complete |
Gemini 3.1 Pro is a preview model (restrictive rate limits). Free tier available for development and small projects. Gemini 3.5 Flash GA May 19 2026 — fast, high-volume tier at 25% less than 3.1 Pro. 76.2% Terminal-Bench 2.1, 83.6% MCP Atlas, 84.2% CharXiv Reasoning. Gemini 3 Flash Preview is a budget-friendly general-purpose model ($0.50/$3.00 per MTok). Gemini 3.1 Pro scores 80.6% on SWE-Bench Verified — competitive with Claude Opus 4.6 (80.8%) and DeepSeek V4 Flash (79%).
DeepSeek
Current as of June 2026. Source: api-docs.deepseek.com
DeepSeek V4 is the current flagship, launched March 2026. 671B total params, 37B active MoE, 1M context. SWE-Bench Verified: V4 Pro Max 80.6%, V4 Flash ~79%. V4 Flash is the default workhorse; V4 Pro standard pricing at $0.435/$0.87 per MTok.
New: DeepSeek V4 Pro Max
Released Apr 24 2026. 1.6T params, 49B active MoE, 1M context, open-weight on HuggingFace. 80.6% SWE-Bench Verified. Available at V4 Pro pricing (same API endpoint).
| Model | Cache Hit Input /1M | Cache Miss Input /1M | Output /1M | Context | Notes |
|---|---|---|---|---|---|
| deepseek-v4-flash | $0.0028 | $0.14 | $0.28 | 1M | Default route. 384K max output |
| deepseek-v4-pro | $0.003625 | $0.435 | $0.87 | 1M | Standard pricing. 1.6T params. 80.6% SWE-Bench Verified |
Cache hit prices reduced to 1/10 of launch price from Apr 26 2026.
Older aliases deepseek-chat and deepseek-reasoner map to V4 Flash (non-thinking / thinking) and retire after Jul 24 2026.
New accounts get 5M free tokens.
Legacy Models
| Model | Input /1M | Output /1M | Cache Hit | Context | Notes |
|---|---|---|---|---|---|
| DeepSeek V3.2 (Chat) | $0.28 | $0.42 | $0.028 | 128K | Previous gen, still available |
| DeepSeek R1 | $0.55 | $2.19 | $0.14 | 128K | Dedicated reasoning model |
DeepSeek V3.2: 73.0% SWE-Bench Verified. R1: chain-of-thought reasoning, ~96% cheaper than OpenAI o1. DeepSeek web chat at chat.deepseek.com is free for individual users.
OpenAI (ChatGPT)
Current as of June 2026. Source: openai.com/api/pricing
GPT-5 Family (Current Flagship)
| Model | Input /1M | Output /1M | Cached Input | Context | Notes |
|---|---|---|---|---|---|
| GPT-5.5 (≤272K) | $5.00 | $30.00 | $0.50 | 1M | 88.7% SWE-Bench Verified (#1), 58.6% SWE-Bench Pro. Flagship reasoning + coding |
| GPT-5.5 (>272K) | $10.00 | $45.00 | $1.00 | 1M | Long context tier >272K tokens |
| GPT-5.5 Pro | $30.00 | $180.00 | — | 1M | Premium tier for research-grade problems |
| GPT-5.4 (≤272K) | $2.50 | $15.00 | $0.25 | 1M | ~80% SWE-Bench Verified. 59.1% SWE-Bench Pro |
| GPT-5.4 (>272K) | $5.00 | $22.50 | $0.50 | 1M | Long context tier >272K tokens |
| GPT-5.4 Mini | $0.75 | $4.50 | $0.075 | 400K | Affordable reasoning. Supports reasoning effort control |
| GPT-5.4 Nano | $0.20 | $1.25 | — | 400K | Fastest, cheapest 5.4 tier. Ideal for summaries, classification |
| GPT-5.3 Codex | $1.75 | $14.00 | — | 400K | 85.0% SWE-Bench Verified (#3). 56.8% SWE-Bench Pro. Coding specialist |
GPT-4.1 Family (Production Workhorse)
| Model | Input /1M | Output /1M | Cached Input | Context | Notes |
|---|---|---|---|---|---|
| GPT-4.1 | $2.00 | $8.00 | $0.50 | 1M | Recommended production model. Strong coding + long context |
| GPT-4.1 Mini | $0.40 | $1.60 | $0.10 | 1M | Good balance of power and affordability |
| GPT-4.1 Nano | $0.10 | $1.40 | — | 1M | Cheapest OpenAI model. Classification, extraction, routing |
o-Series (Reasoning Models)
| Model | Input /1M | Output /1M | Cached Input | Context | Notes |
|---|---|---|---|---|---|
| o4-mini | $1.10 | $4.40 | $0.275 | 200K | Best-value reasoning. Math, science, complex logic |
| o3 | $2.00 | $8.00 | — | — | Flagship reasoning. Chain-of-thought built in |
Batch API saves 50% on all models. Prompt caching discounts: up to 90% off (GPT-5.5), 75% off (GPT-4.1). GPT-5.5 scores 88.7% SWE-Bench Verified and 58.6% SWE-Bench Pro. GPT-5.4 scores ~80% SWE-Bench Verified and 59.1% SWE-Bench Pro (xHigh). GPT-5.5 Pro tier ($30/$180) is available for research-grade problems. GPT-5.3 Codex scores 85.0% SWE-Bench Verified and 56.8% SWE-Bench Pro. GPT-4.1 is OpenAI’s recommended production default for most workloads.
MiniMax
Current as of June 2026. Source: platform.minimax.io, OpenRouter
Coding Models
| Model | Input /1M | Output /1M | Context | Max Output | SWE-Bench | Speed | Notes |
|---|---|---|---|---|---|---|---|
| M3 (≤512K) | $0.30 (permanent 50% off, was $0.60) | $1.20 (permanent 50% off, was $2.40) | 512K+ | — | — | ~50 TPS | Permanent 50% off. $0.06/M cached read |
| M3 (>512K) | $0.60 (permanent 50% off, was $1.20) | $2.40 (permanent 50% off, was $4.80) | 512K+ | — | — | — | Long-context tier. Permanent 50% off |
| M2.7 | $0.30 | $1.20 | 205K | 131K | — | ~80 TPS | Released Mar 18 2026. $0.06/M cached read |
| M2.5 Standard | $0.30 | $1.20 | 256K | — | 80.2% | ~50 TPS | Legacy. Automatic cache (no config) |
| M2.5 Lightning | $0.60 | $2.40 | 256K | — | 80.2% | ~100 TPS | Legacy. Priority scheduling |
M2.5 Standard: Still one of the best value coding models at $0.30/$1.20. Near Claude Opus 4.6 (80.8%). M3: Latest generation. $0.30/$1.20 (≤512K, permanent 50% off list $0.60/$2.40). $0.60/$2.40 (>512K, also permanent 50% off list $1.20/$4.80). Supports >512K context via Priority tier. OpenCode Go estimates: M2.5 ~6,300 req/5h (legacy price), M2.7 ~3,400 req/5h, M3 ~3,000 req/5h.
Subscription Plans
| Plan | Price | Description |
|---|---|---|
| Token Plan | Subscription | Quotas for individual builders and Teams |
| Credits | Prepaid | Same resource coverage as Token Plan |
| Pay-as-you-go | Per-token | Standard API endpoint billing |
Qwen (Alibaba)
Current as of June 2026. Source: DashScope direct pricing
Current Gen (Qwen3.7 & Qwen3.6)
Qwen3.7 (Latest — May 2026)
| Model | Input /1M | Output /1M | Context | Notes |
|---|---|---|---|---|
| Qwen3.7 Max | $2.50 | $7.50 | 1M | May 20 2026. Closed-weights text flagship. 1M context. Intelligence Index v4.0 #5 overall. Lowest hallucination rate among frontier models (22.9%). Anthropic API protocol native. 35h autonomous operation. Also available via Novita at $1.25/$3.75 (third-party). |
| Qwen3.7 Plus | TBA | TBA | 1M | Multimodal (vision + text) variant. Announced May 21 2026. |
Qwen3.7 Max (May 20-21 2026): Alibaba’s newest flagship, announced at Alibaba Cloud Summit. $2.50/$7.50 per MTok — ~6x cheaper than Claude Opus 4.7. 1M context. Terminal-Bench Hard: 50.8%. MCP-Atlas: 76.4. Anthropic API protocol support means it works as a drop-in Claude Code backend. Open-weight variants expected June/July 2026 following Qwen3.6 release pattern.
Qwen3.6
| Model | Input /1M | Output /1M | Context | SWE-Bench | Notes |
|---|---|---|---|---|---|
| Qwen3.6 Plus | $0.325 | $1.95 | 1M | 78.8% Verified | Apr 2 2026. Hybrid attention + MoE. Reasoning by default |
| Qwen3.6 Flash | $0.25 | $1.50 | 1M | — | Cost-optimized tier |
| Qwen3.6 Max Preview | $1.30 | $7.80 | 256K | SWE-Bench Pro #1 | Apr 20 2026. Closed-weights flagship. Leads SWE-Bench Pro, Terminal-Bench 2.0, SkillsBench, SciCode |
Qwen3.6 Plus: 78.8% SWE-Bench Verified — within 2 points of Claude Opus 4.6 (80.8%) at 1/30th the input price. 1M native context, 65K max output. Reasoning enabled by default (no mode toggle). Qwen3.6-27B (dense, Apache 2.0): 77.2% SWE-Bench Verified — strong self-hosting option. Qwen3.6-Max-Preview (Apr 20 2026): First closed-weights Qwen flagship. $1.30/$7.80 per MTok. 256K context. 57.3% SWE-Bench Pro — surpasses GPT-5.3-Codex (56.8%) and MiMo-V2.5-Pro (57.2%). Tops SWE-Bench Pro + 5 other coding benchmarks (Terminal-Bench 2.0, SkillsBench, QwenClawBench, QwenWebBench, SciCode) at launch.
Previous Gen (Qwen3.5)
| Model | Input /1M | Output /1M | Context | Notes |
|---|---|---|---|---|
| Qwen3.5 Plus | $0.40 ($0.115 Global) | $2.40 ($0.688 Global) | 1M | Feb 2026 release. 65K max output. Thinking mode: output at same price |
| Qwen3.5 397B A17B | Free | Free | 262K | Open-weight MoE flagship |
Qwen-Max (Legacy Flagship)
| Model | Input /1M | Output /1M | Context |
|---|---|---|---|
| qwen3-max (0-32K) | $1.20 | $6.00 | 252K |
| qwen3-max (32K-128K) | $2.40 | $12.00 | 252K |
| qwen3-max (128K-252K) | $3.00 | $15.00 | 252K |
| qwen-max (older) | $1.60 | $6.40 | — |
All Qwen models support native tool-calling, JSON-mode, and OpenAI-compatible API shapes. Batch calling: 50% off. Context caching discounts available on supported models.
GLM / Z.ai
Current as of June 2026. Source: docs.z.ai
Flagship Models (GLM-5 Series)
| Model | Context | SWE-Bench | Input /1M | Output /1M | Cached Input | License |
|---|---|---|---|---|---|---|
| GLM-5.2 | 1M | — | $1.40 | $4.40 | $0.26 | Proprietary |
| GLM-5.1 | 203K | Pro 58.4% (best-in-class) | $1.40 | $4.40 | $0.26 | MIT, 754B params |
| GLM-5 | 202K | Verified 77.8% | $1.00 | $3.20 | $0.20 | MIT, 744B/40B MoE |
| GLM-5-Turbo | 202K | — | $1.20 | $4.00 | $0.24 | Proprietary |
GLM-5.1 (Apr 7 2026): 8-hour autonomous runs, 1,700 agentic steps. Surpasses GPT-5.4 and Claude Opus 4.6 on SWE-Bench Pro. GLM-5: 744B params, 40B active MoE, 28.5T token pretraining. GLM-5.2 (Jun 2026): Latest flagship. 1M context, 128K max output. Long-horizon engineering specialist. Same pricing as GLM-5.1 ($1.40/$4.40).
Previous Gen (GLM-4 Series)
| Model | Context | Input /1M | Output /1M | Cached Input | Notes |
|---|---|---|---|---|---|
| GLM-4.7 | 128K | $0.60 | $2.20 | $0.11 | 73.8% SWE-Bench Verified |
| GLM-4.7-FlashX | 203K | $0.07 | $0.40 | $0.01 | Fast inference variant |
| GLM-4.6 | 128K | $0.60 | $2.20 | $0.11 | Previous generation |
| GLM-4.5-X | 128K | $2.20 | $8.90 | $0.45 | Premium tier |
| GLM-4.5 | 128K | $0.60 | $2.20 | $0.11 | Standard tier |
| GLM-4.5-Air | 128K | $0.20 | $1.10 | $0.03 | Lightweight, Haiku-class |
| GLM-4.5-AirX | 128K | $1.10 | $4.50 | $0.22 | Fast Air variant |
| GLM-4-32B-0414-128K | 128K | $0.10 | $0.10 | — | Budget open-weight |
Free Models
| Model | Context | Input | Output |
|---|---|---|---|
| GLM-4.7-Flash | 203K | Free | Free |
| GLM-4.5-Flash | — | Free | Free |
Vision Models
| Model | Input /1M | Output /1M | Cached Input |
|---|---|---|---|
| GLM-5V-Turbo | $1.20 | $4.00 | $0.24 |
| GLM-4.6V | $0.30 | $0.90 | $0.05 |
| GLM-4.6V-FlashX | $0.04 | $0.40 | $0.004 |
| GLM-OCR | $0.03 | $0.03 | — |
| GLM-4.6V-Flash | Free | Free | Free |
Xiaomi MiMo
Current as of June 2026. V2 launched Mar 18 2026, V2.5 launched Apr 22 2026. Source: mimo-v2.com
| Model | Input /1M | Output /1M | Context | Modalities | Notes |
|---|---|---|---|---|---|
| MiMo-V2-Pro (≤256K) | $1.00 | $3.00 | 1M | Text | 78.0% SWE-Bench. 1T params, 42B active |
| MiMo-V2-Pro (256K–1M) | $2.00 | $6.00 | 1M | Text | Long-context tier |
| MiMo-V2.5 (≤256K) | $0.40 ($0.08 cached) | $2.00 | 1M | Text, Image, Audio, Video | Apr 22 2026. Native omnimodal. 1M context. OpenAI-compatible |
| MiMo-V2.5 (256K–1M) | $0.80 | $4.00 | 1M | Text, Image, Audio, Video | Long-context tier |
| MiMo-V2.5-Pro (≤256K) | $1.00 ($0.20 cached) | $3.00 | 1M | Text | Apr 22 2026. MIT license. 1T params. 57.2% SWE-Bench Pro |
| MiMo-V2.5-Pro (256K–1M) | $2.00 | $6.00 | 1M | Text | Long-context tier |
| MiMo-V2-Omni | ~$1.00 | ~$3.00 | 256K | Text, Image, Audio, Video | Multimodal flagship |
| MiMo-V2-Flash | $0.10 | $0.30 | 256K | Text | Open-source foundation model |
| MiMo-V2-TTS | Free | Free | — | Audio | Limited time promo |
API at platform.xiaomimimo.com. OpenAI-compatible. Credit plans available: Lite $6/mo, Standard $16/mo, Pro $50/mo, Max $100/mo.
Kimi / Moonshot AI (K2.7 Code · K2.6)
Current as of June 2026. Source: kimi.com, K2.7 Code, platform.kimi.ai, OpenRouter
All three models: 1T params, 32B active MoE, 384 experts, MIT license. K2.7 Code is a coding-specific variant (thinking mode required).
| Model | Cache Hit /1M | Cache Miss /1M | Output /1M | Context | SWE-Bench |
|---|---|---|---|---|---|
| kimi-k2.7-code | $0.19 | $0.95 | $4.00 | 262K | — |
| kimi-k2.6 | $0.16 | $0.95 | $4.00 | 262K | Verified 80.2%, Pro 58.6%, BrowseComp 83.2% |
| kimi-k2.5 | — | $0.40 | $1.90 | 256K | Verified 76.8%, BrowseComp 78.4% |
K2.7 Code: Coding-specific agentic model. ~30% less thinking tokens vs K2.6. MCP Atlas 76.0, MCP Mark Verified 81.1. Available via Kimi Code (default model) and Kimi API. See kimi.com.
K2.6: 300 parallel sub-agents, 4,000+ tool calls, 12+ hr continuous execution. K2.5: 100 parallel sub-agents.
Membership Plans
| Plan | Price/mo | Agent Usage |
|---|---|---|
| Adagio | Free | 6 |
| Moderato | $15 | 60 |
| Allegretto | $31 | 150 |
| Allegro | $79 | 360 |
| Vivace | $159 | 720 |
Mistral (Medium 3.5)
Current as of June 2026. Source: la-plateforme.mistral.ai, tokencost.app
| Model | Input /1M | Output /1M | Context | Max Output | SWE-Bench Verified | Notes |
|---|---|---|---|---|---|---|
| Mistral Medium 3.5 | $1.50 | $7.50 | 256K | 33K | 77.6% | 128B dense, multimodal. Released Apr 29 2026. |
Mistral Medium 3.5 consolidates Medium 3.1 + Magistral + Devstral 2 into one unified model. Modified MIT license with open weights. 77.6% SWE-Bench Verified — competitive with Qwen3.6-27B (77.2%) and ahead of DeepSeek V3.2 (73.0%).
xAI (Grok)
Current as of June 2026. Source: docs.x.ai, xAI pricing
| Model | Input /1M | Output /1M | Cached Input | Context | SWE-Bench | Notes |
|---|---|---|---|---|---|---|
| Grok 4.3 | $1.25 | $2.50 | — | 1M | ~72-75% (self-reported) | Launched Apr 30 2026. Current flagship. 1M context. |
| Grok 4.20 | $1.25 | $2.50 | $0.20 | 1M | — | 85% cached-input discount. Retired slugs (Grok 4, Grok 4.1 Fast) redirect here. |
| Grok 4 | $3.00 | $15.00 | — | 256K | 72-75% (self-reported) | Original Grok 4 flagship. Deprecated May 15 2026 — now redirects to Grok 4.3. |
| Grok Build 0.1 | $1.00 | $2.00 | — | 256K | — | Coding specialist. 20% cheaper than Grok 4.3. |
Grok 4.3 self-reports ~72-75% on SWE-Bench Verified (independent testing with SWE-agent scaffold: 58.6%). Aider Polyglot: 79.6%. xAI retired Grok 4, Grok 4.1 Fast, and Grok Code Fast 1 slugs on May 15, 2026 — these now redirect to Grok 4.3 at $1.25/$2.50 pricing. xAI offers up to $150/month free API credits via data-sharing program. Unique real-time X search grounding.
OpenCode Go
Source: docs.openclaw.ai. Dollar-value limits ($12/5h, $30/week, $60/month).
Available Models
| Model Ref | Name |
|---|---|
| opencode-go/glm-5 | GLM-5 |
| opencode-go/glm-5.1 | GLM-5.1 |
| opencode-go/glm-5.2 | GLM-5.2 |
| opencode-go/kimi-k2.5 | Kimi K2.5 |
| opencode-go/kimi-k2.6 | Kimi K2.6 (3x limits) |
| opencode-go/kimi-k2.7-code | Kimi K2.7 Code |
| opencode-go/deepseek-v4-pro | DeepSeek V4 Pro |
| opencode-go/deepseek-v4-flash | DeepSeek V4 Flash |
| opencode-go/mimo-v2-omni | MiMo V2 Omni |
| opencode-go/mimo-v2-pro | MiMo V2 Pro |
| opencode-go/mimo-v2.5 | MiMo V2.5 |
| opencode-go/mimo-v2.5-pro | MiMo V2.5 Pro |
| opencode-go/minimax-m2.5 | MiniMax M2.5 |
| opencode-go/minimax-m2.7 | MiniMax M2.7 |
| opencode-go/minimax-m3 | MiniMax M3 |
| opencode-go/qwen3.5-plus | Qwen3.5 Plus |
| opencode-go/qwen3.6-plus | Qwen3.6 Plus |
| opencode-go/qwen3.6-max-preview | Qwen3.6 Max Preview |
| opencode-go/mistral-medium-3.5 | Mistral Medium 3.5 |
| opencode-go/grok-4.3 | Grok 4.3 |
| opencode-go/grok-4.20 | Grok 4.20 |
Request Estimates (June 2 2026)
| Model | Per 5h | Per Week | Per Month |
|---|---|---|---|
| GLM-5.1 | 880 | 2,150 | 4,300 |
| GLM-5 | 1,150 | 2,880 | 5,750 |
| Kimi K2.5 | 1,850 | 4,630 | 9,250 |
| MiMo-V2-Pro | 1,290 | 3,225 | 6,450 |
| MiMo-V2.5 | 1,290 | 3,225 | 6,450 |
| MiMo-V2.5-Pro | 1,290 | 3,225 | 6,450 |
| MiMo-V2-Omni | 2,150 | 5,450 | 10,900 |
| Qwen3.6 Plus | 3,300 | 8,200 | 16,300 |
| Qwen3.6 Max Preview | 820 | 2,050 | 4,100 |
| MiniMax M3 | ~3,000 | ~7,500 | 15,000 |
| MiniMax M2.7 | 3,400 | 8,500 | 17,000 |
| MiniMax M2.5 | 3,150 | 7,950 | 15,900 |
| Mistral Medium 3.5 | ~2,500 | ~6,200 | 12,500 |
| Qwen3.5 Plus | 10,200 | 25,200 | 50,500 |
MiniMax M2.5: 80.2% SWE-Bench — near Claude Opus 4.6 (80.8%). DeepSeek V4 Pro: Standard pricing at $0.435/$0.87 per MTok.
Notes
- BytePlus ModelArk: Quota shared across Claude Code, Cursor, Cline, Codex CLI, Kilo Code, Roo Code, OpenCode
- GitHub Copilot: Premium requests shared across all features; extra $0.04 each on Pro/Pro+
- Claude Code: Exact request counts not published — only relative multipliers
- GLM quota multipliers: Peak hours drain 3x quota; off-peak 2x; GLM-4.7/4.5-Air always 1x
- MiMo: Pure credit pool, no 5h/windows, credits expire month-end
- Kimi: API billed separately — not included in membership
Benchmark Note: SWE-Bench Verified measures a model’s ability to resolve real-world GitHub issues from code repositories. Not all providers publish scores — the chart above only includes models with verified data.