OpenAI Codex
newGPT-5.5
Long-horizon coding agents, refactors, debugging, tool-heavy work.
- Context
- Released to ChatGPT, Codex, and the API after the Apr 24 update.
- Benchmark signal
- Terminal-Bench 2.0: 82.7%
- Pricing signal
- Check API plan/pricing
AI model radar
Know which model to build with before you ship.
A builder-facing comparison room for Claude Code, OpenAI Codex, Gemini, and the other agentic models makers actually use.
Claude Code
frontierClaude Opus 4.7
Hard software engineering, review, long-running agents, visual UI work.
- Context
- Generally available across Claude products, API, Bedrock, Vertex AI, and Foundry.
- Benchmark signal
- CursorBench: 70% reported
- Pricing signal
- $5 / $25 per 1M tokens
Kimi Code
openKimi K2.6
Open-source long-horizon coding, devops, performance work, agent swarms.
- Context
- Latest Kimi model, available through Kimi.com, API, app, and Kimi Code.
- Benchmark signal
- Terminal-Bench 2.0 + SWE-Bench Pro signals
- Pricing signal
- API / open-source deployment
Qwen Code
localQwen3-Coder-Next
Self-hosted coding agents, local dev loops, tool calling, broad language support.
- Context
- 80B total / 3B active open-weight coding model with 256K context.
- Benchmark signal
- Agent-centric SWE-Bench + Terminal-Bench focus
- Pricing signal
- Open weights / provider API
DeepSeek
valueDeepSeek V4-Pro
Cost-effective long-context agents, tool use, open-weight coding workflows.
- Context
- Preview release with 1M context; Pro is 1.6T total / 49B active params.
- Benchmark signal
- Open-source SOTA claim on agentic coding benchmarks
- Pricing signal
- $0.435 / $0.87 per 1M tokens during launch discount
Google Gemini
multimodalGemini 3 Pro Preview
Multimodal reasoning, autonomous coding, large-context app workflows.
- Context
- Preview model with 1M input and 64k output token limits.
- Benchmark signal
- Gemini 3 series: agentic coding + thinking
- Pricing signal
- $2 / $12 per 1M tokens under 200k
Claude Sonnet
scaleClaude Sonnet 4.6
Daily coding, product work, agent planning, high-volume engineering teams.
- Context
- Most capable Sonnet tier, with 1M context in beta.
- Benchmark signal
- Strong coding and agent upgrade over Sonnet 4.5
- Pricing signal
- $3 / $15 per 1M tokens
Mistral Code
enterpriseDevstral + Codestral 25.08
Private coding stacks, inline completion, repo search, and enterprise agents.
- Context
- Codestral handles low-latency FIM completion; Devstral powers multi-step development.
- Benchmark signal
- Devstral Medium: 61.6% SWE-Bench Verified
- Pricing signal
- API / enterprise / self-hosted stack
benchmarks
What the page should help makers compare
Terminal-Bench 2.0 from OpenAI's launch table.
Kimi's latest open-source coder emphasizes long-running execution and agent swarms.
Qwen3-Coder-Next targets coding agents with 80B total / 3B active compute.
Strong Sonnet tier for coding, agents, and professional work at scale.
DeepSeek V4 Preview brings 1M context and agentic coding focus to open-weight workflows.
Google's preview model for multimodal reasoning and autonomous coding.
decision matrix
Pick by job, not hype.
Use caseLikely first pickWhy
Autonomous coding sprintGPT-5.5 or Opus 4.7Strong long-horizon coding and tool-use signals.
Daily maker workflowSonnet 4.6Balanced frontier capability, scale, and 1M context beta.
Multimodal app reasoningGemini 3 ProLarge multimodal context with thinking and code execution support.
Cost-sensitive iterationHaiku / Flash classUse fast tiers for repeated edits, tests, and small tasks.
source links