FR

Model comparison

About thirty AI models for coding and agents side by side (Anthropic, OpenAI, Google, Mistral, Qwen, DeepSeek, Kimi, GLM…): API list price, announced coding score, context, license and release date. Filter by vendor or open weights, click a column header to sort.

30 models, 6 of them from directory entries.

Model comparison. Column headers are sort buttons.
Sources
Clauderef.: Claude Sonnet 5.5Anthropic$2$10—1,000,000—Noprice — Claude · context — Claude
GPTref.: GPT-6.1 SolOpenAI$2above 272K input tokens: 2x input, 1.5x output$10—1,050,000—No—price — GPT · context — GPT
Muse Spark Openref.: Muse Spark Open (816B-A42B)Meta—no first-party price: API via third-party hosts—60%DeepSWE v1.1 · announced by Meta, preferred harness, xhigh reasoning1,000,000Muse Model LicenseYesscore — Muse Spark Open · context — Muse Spark Open
MiniMax M3.1 FlashMiniMax$0.15flat rate; $0.015 cached input$0.673.7%DeepSWE v1.1 · announced by MiniMax1,000,000—Noprice — MiniMax M3.1 Flash · score — MiniMax M3.1 Flash · context — MiniMax M3.1 Flash
DeepSeek V4.1-Flash-1017DeepSeek$0.3peak hours, cache miss; half price off-peak$1.277%DeepSWE v1.1 · announced by DeepSeek, Mini-SWE-V2 harness1,000,000MITYesprice — DeepSeek V4.1-Flash-1017 · score — DeepSeek V4.1-Flash-1017 · context — DeepSeek V4.1-Flash-1017
Gemini 4 ArgonGoogle$2launch price until Nov 30, 2026, then $4/$20$1077.9%DeepSWE v1.1 · announced by Google1,000,000—Noprice — Gemini 4 Argon · score — Gemini 4 Argon · context — Gemini 4 Argon
Claude Fable 5.1Anthropic$10$50—1,000,000—Noprice — Claude Fable 5.1 · context — Claude Fable 5.1 · release — Claude Fable 5.1
Claude Opus 5.5Anthropic$4$20—1,000,000—Noprice — Claude Opus 5.5 · context — Claude Opus 5.5 · release — Claude Opus 5.5
Claude Haiku 5.5Anthropic$0.1prompts ≤ 100K tokens; $0.50/$2.50 above$0.5—1,000,000—Noprice — Claude Haiku 5.5 · context — Claude Haiku 5.5 · release — Claude Haiku 5.5
GPT-6 AstraOpenAI$10above 272K input tokens: 2x input, 1.5x output$50—1,050,000—Noprice — GPT-6 Astra · context — GPT-6 Astra · release — GPT-6 Astra
GPT-6 LunaOpenAI$0.1$0.5—1,050,000—Noprice — GPT-6 Luna · context — GPT-6 Luna · release — GPT-6 Luna
GPT-5.3-CodexOpenAI$1.75specialized Codex model$14—400,000—Noprice — GPT-5.3-Codex · context — GPT-5.3-Codex · release — GPT-5.3-Codex
gpt-oss-120bOpenAI—not listed on the OpenAI API pricing page——131,072Apache 2.0Yes—context — gpt-oss-120b · license — gpt-oss-120b
Gemini 3.8 FlashGoogle$0.75through Dec 31, 2026, then $1.50/$7.50$3.75—1,048,576—Noprice — Gemini 3.8 Flash · context — Gemini 3.8 Flash · release — Gemini 3.8 Flash
Gemini 3.5 Flash-LiteGoogle$0.3$2.5—1,048,576—Noprice — Gemini 3.5 Flash-Lite · context — Gemini 3.5 Flash-Lite · release — Gemini 3.5 Flash-Lite
Gemma 4 31BGoogle—free tier of the Gemini API only, no paid price——256,000Apache 2.0Yescontext — Gemma 4 31B · release — Gemma 4 31B · license — Gemma 4 31B
Grok 4.7xAI$2$671%DeepSWE v1.1 · high effort, announced by xAI500,000—Noprice — Grok 4.7 · score — Grok 4.7 · context — Grok 4.7 · release — Grok 4.7
Mistral Large 4Mistral AI$1.36public preview$4.18—1,000,000——price — Mistral Large 4 · context — Mistral Large 4 · release — Mistral Large 4
Mistral Medium 3.5Mistral AI$1.5$7.5—256,000Modified MITYesprice — Mistral Medium 3.5 · context — Mistral Medium 3.5 · release — Mistral Medium 3.5 · license — Mistral Medium 3.5
Mistral Small 4Mistral AI$0.15$0.6—256,000Apache 2.0Yesprice — Mistral Small 4 · context — Mistral Small 4 · release — Mistral Small 4 · license — Mistral Small 4
Codestral 25.08Mistral AI$0.3code completion (FIM)$0.9—128,000——price — Codestral 25.08 · context — Codestral 25.08 · release — Codestral 25.08
GLM-5.3Z.ai$1.4$4.466.9%DeepSWE v1.1 · announced by Z.ai1,000,000GLM-5.3 LicenseGLM-5.3 License (MIT-style; Z.ai security review for API providers above $10B revenue over 12 months)Yesprice — GLM-5.3 · score — GLM-5.3 · context — GLM-5.3 · release — GLM-5.3 · license — GLM-5.3
GLM-5.3-FlashZ.ai$0.15$0.563.4%DeepSWE v1.1 · announced by Z.ai1,000,000MITYesprice — GLM-5.3-Flash · score — GLM-5.3-Flash · context — GLM-5.3-Flash · release — GLM-5.3-Flash · license — GLM-5.3-Flash
Kimi K3Moonshot AI$3cache-miss input; $0.30 cache hit$1567.3%DeepSWE v1.1 · mini-SWE-agent harness (DeepSWE leaderboard), cited by Moonshot1,000,000Kimi K3 LicenseKimi K3 License (MIT-style; separate agreement for API providers above $20M revenue over 12 months, “Kimi K3” attribution above 100M monthly users or $20M monthly revenue)Yes—price — Kimi K3 · score — Kimi K3 · context — Kimi K3 · license — Kimi K3
DeepSeek V4-Pro-0813DeepSeek$1.32peak hours, cache miss; half price off-peak$3.9662.7%DeepSWE v1.1 · DeepSeek Harness (minimal mode), max effort, announced by DeepSeek1,000,000MITYesprice — DeepSeek V4-Pro-0813 · score — DeepSeek V4-Pro-0813 · context — DeepSeek V4-Pro-0813 · release — DeepSeek V4-Pro-0813 · license — DeepSeek V4-Pro-0813
MiniMax M3MiniMax$0.3input ≤ 512K tokens; $0.60/$2.40 above$1.2—1,000,000MiniMax Community LicenseMiniMax Community License (non-commercial use)Yesprice — MiniMax M3 · context — MiniMax M3 · release — MiniMax M3 · license — MiniMax M3
Qwen3.8-MaxAlibaba (Qwen)$2international (Singapore) price$656.6%DeepSWE v1.1 · best of Claude Code / mini-SWE-agent, announced by Qwen1,000,000Qwen3.8-Max LicenseQwen3.8-Max License (separate licence for API services and work assistants above $50M annual revenue)Yesprice — Qwen3.8-Max · score — Qwen3.8-Max · context — Qwen3.8-Max · release — Qwen3.8-Max · license — Qwen3.8-Max
Qwen3.8-27BAlibaba (Qwen)$0.5international (Singapore) price$342.2%DeepSWE v1.1 · Claude Code harness, announced by Qwen262,144Apache 2.0Yes—price — Qwen3.8-27B · score — Qwen3.8-27B · context — Qwen3.8-27B · license — Qwen3.8-27B
Qwen3.8-FlashAlibaba (Qwen)$0.15international (Singapore) price$0.47—————price — Qwen3.8-Flash
Command A+Cohere———128,000Apache 2.0Yes—context — Command A+ · license — Command A+

Standard list prices in US dollars per million tokens, excluding caching, discounts and long-context tiers. DeepSWE v1.1 scores are announced by each vendor under its own test conditions (agent harness, reasoning effort): they are not strictly comparable. Context windows announced as “k” or “M” are converted to thousands or millions of tokens. “—” = not published or not verified. · Verified

Comparisons · Glossary · Tool directory