Every model on this page participates in the same account-level discount ladder: save 25% from signup and up to 30% as usage grows. The catalog is organized by family and provider, while benchmark coverage stays conservative and avoids guessing from nearby variants.
Grouped by model family and provider, with GLM 5.2 and DeepSeek V4 prioritized ahead of Gemini.
3 models
Code-focused GPT model
A coding-oriented GPT tier aimed at repository edits, software engineering workflows, and agent loops.
Flagship GPT model for general reasoning
A frontier GPT model aimed at stronger reasoning, instruction following, and general assistant work.
Smaller GPT model for higher-throughput usage
A lighter GPT option suited to lower-cost traffic, background tasks, and simpler coding assistance.
3 models
Fast Claude model for lightweight tasks
A lighter Claude option suited to fast responses, routine assistant turns, and lower-cost background workflows.
High-end Claude model for difficult reasoning
A flagship Claude tier aimed at harder reasoning, review-heavy workflows, and higher-accuracy engineering tasks.
Balanced Claude model for daily engineering work
A more balanced Claude tier for everyday coding, assistant flows, and mixed reasoning workloads.
2 models
Multilingual general assistant model
A GLM family model that fits multilingual assistant flows and general application integrations.
Advanced multilingual reasoning and coding model
A current GLM family option for reasoning-heavy chat, coding, tool orchestration, and multilingual workflows.
2 models
Fast, cost-aware DeepSeek V4 model
The faster DeepSeek V4 option for reasoning, coding, automation, and throughput-sensitive production workloads.
Higher-capability DeepSeek V4 model
The stronger DeepSeek V4 option for complex reasoning, coding, agents, and long-context production workloads.
1 models
Reasoning-oriented general model
A Kimi family model with public software engineering benchmark coverage and interest for longer-context workflows.
1 models
Reasoning and coding model
A MiniMax family model with one of the stronger publicly visible SWE-bench Verified signals in this catalog.
3 models
Fast-response Qwen model
A Qwen family variant positioned for lower latency and lighter assistant or coding flows.
Balanced Qwen model
A balanced Qwen tier positioned between flash variants and heavier reasoning options.
Higher-capability Qwen model
A newer Qwen option intended for stronger reasoning and richer enterprise assistant workloads.
3 models
Fast Gemini preview model
A preview Gemini tier oriented toward fast response and interactive assistant workloads.
Lightweight Gemini preview model
A lighter preview Gemini option suited to lower-cost throughput and simple assistant turns.
Higher-capability Gemini preview model
A preview Gemini tier intended for stronger reasoning and more demanding assistant or coding tasks.
The table below favors exact public evidence over aggressive family-level mapping. That keeps the page honest even when model naming evolves faster than public benchmark rows.
| Model | Provider | Coding benchmarks |
|---|---|---|
claude-haiku-4.5 Fast Claude model for lightweight tasks | Anthropic | No exact public score confirmed in the current checked sources. |
claude-opus-4.6 High-end Claude model for difficult reasoning | Anthropic | SWE-bench Verified75.60 % resolved Snapshot: 2026-02-17 Leaderboard display name: Claude Opus 4.6 View source |
claude-sonnet-4.6 Balanced Claude model for daily engineering work | Anthropic | SWE-bench VerifiedClaude 4.5 Sonnet (high reasoning) 71.40 % resolved Snapshot: 2026-02-17 Nearby family reference only. Public coverage was found for Claude 4.5 Sonnet, not claude-sonnet-4.6. View source |
deepseek-v4-flash Fast, cost-aware DeepSeek V4 model | DeepSeek | No exact public score confirmed in the current checked sources. |
deepseek-v4-pro Higher-capability DeepSeek V4 model | DeepSeek | No exact public score confirmed in the current checked sources. |
gemini-3-flash-preview Fast Gemini preview model | No exact public score confirmed in the current checked sources. | |
gemini-3.1-flash-lite-preview Lightweight Gemini preview model | No exact public score confirmed in the current checked sources. | |
gemini-3.1-pro-preview Higher-capability Gemini preview model | No exact public score confirmed in the current checked sources. | |
glm-4.7 Multilingual general assistant model | Zhipu AI | OpenCompass69.3 Code score Snapshot: 2026-02-11 CompassBench LLM leaderboard row: GLM-4.7 Open Source - Zhipu AI. This is the leaderboard's Code dimension, not SWE-bench resolved rate. View source |
glm-5.2 Advanced multilingual reasoning and coding model | Zhipu AI | No exact public score confirmed in the current checked sources. |
gpt-5.3-codex Code-focused GPT model | OpenAI | SWE-bench VerifiedGPT-5-2 Codex 72.80 % resolved Snapshot: 2026-02-19 Nearby family reference only. Public coverage was found for GPT-5-2 Codex, not gpt-5.3-codex. View source |
gpt-5.4 Flagship GPT model for general reasoning | OpenAI | SWE-bench VerifiedGPT-5-2 (high reasoning) 72.80 % resolved Snapshot: 2026-02-17 Nearby family reference only. Public coverage was found for GPT-5-2, not gpt-5.4. View source |
gpt-5.4-mini Smaller GPT model for higher-throughput usage | OpenAI | SWE-bench VerifiedGPT-5 Mini 56.20 % resolved Snapshot: 2026-02-17 Nearby family reference only. Public coverage was found for GPT-5 Mini, not gpt-5.4-mini. View source |
kimi-k2.5 Reasoning-oriented general model | Moonshot AI | SWE-bench Verified70.80 % resolved Snapshot: 2026-02-17 Leaderboard display name: Kimi K2.5 (high reasoning) View source |
MiniMax-M2.5 Reasoning and coding model | MiniMax | SWE-bench Verified75.80 % resolved Snapshot: 2026-02-17 Leaderboard display name: MiniMax M2.5 (high reasoning) View source |
qwen3.5-flash Fast-response Qwen model | Alibaba Qwen | No exact public score confirmed in the current checked sources. |
qwen3.5-plus Balanced Qwen model | Alibaba Qwen | No exact public score confirmed in the current checked sources. |
qwen3.6-plus Higher-capability Qwen model | Alibaba Qwen | No exact public score confirmed in the current checked sources. |
These are the public sources used to determine whether an exact benchmark row can be shown.
Source 1
SWE-bench Verified
SWE-bench Verified official leaderboard for exact public rows and resolved-rate percentages.
Source 2
Aider public leaderboards
Aider public coding leaderboards for additional coding signals, but only exact current-model rows should be copied into production content.
Source 3
LiveCodeBench
LiveCodeBench public leaderboard as an additional source when a clear public row is available for the same model naming.
Source 4
OpenCompass
OpenCompass CompassBench LLM leaderboard when an exact model row exposes an explicit Code dimension that can be cited directly.
If you need a model that is not listed on this page, contact us at [email protected].