AI model rankings
Foundation models (LLMs) ranked by performance, public reviews and recency, based on public sources and refreshed regularly.
Showing the top 20 of 81 models.
- #1Claude Opus 5.5Anthropic's current Opus model (September 2026), which Anthropic says performs at the level of Claude Fable 5.1 on most work at a lower price than Opus 5.Tier S, score 94.5$4 input / $20 output per 1M tokens (API); included in Claude Pro, Max, Team and Enterprise plans$4 input / $20 output per 1M tokens (API); included in Claude Pro, Max, Team and Enterprise plans
- #2GPT-6 AstraOpenAI's GPT-6 frontier model (September 2026) for end-to-end reasoning, coding, computer use and research, with a 1.05M-token context window.Tier S, score 93.5API $10 input / $50 output per 1M tokensAPI $10 input / $50 output per 1M tokens
- #3Claude Sonnet 5.5Anthropic's faster, lower-cost Claude 5.5 model for coding, agents, and everyday knowledge work.Tier S, score 92.9$2 input / $10 output per 1M tokens (API); cache reads $0.20 per 1M$2 input / $10 output per 1M tokens (API); cache reads $0.20 per 1M
- #4Gemini 4 ArgonNEWGoogle's flagship Gemini 4 reasoning model (September 30, 2026) with a 1M-token context window and very long outputs, in the Gemini app and API.Tier S, score 92.2API $2 input / $10 output per 1M tokens through Nov 30, 2026 ($4 / $20 from Dec 1, 2026)API $2 input / $10 output per 1M tokens through Nov 30, 2026 ($4 / $20 from Dec 1, 2026)
- #5GPT-6.1 SolOpenAI's September 2026 Sol update for coding, computer use and professional work, positioned close to GPT-6 Astra at a fraction of its token price.Tier S, score 91.8API $2 input / $10 output per 1M tokens ($0.10 cached input)API $2 input / $10 output per 1M tokens ($0.10 cached input)
- #6Kimi K3.1 ProNEWMoonshot AI's flagship Kimi model (October 2026) for long-horizon coding and knowledge work, with native vision and a 1M-token context window.Tier S, score 91.6API about $2 input / $10 output per 1M tokens via the Kimi API (per Artificial Analysis); open weights due by Nov 4, 2026API about $2 input / $10 output per 1M tokens via the Kimi API (per Artificial Analysis); open weights due by Nov 4, 2026
- #7Claude Fable 5.1Anthropic's Fable 5 point release, tied at the top of the Artificial Analysis Intelligence Index.Tier S, score 91.1API $10 input / $50 output per 1M tokens; Claude Pro and Max via usage credits, Team and EnterpriseAPI $10 input / $50 output per 1M tokens; Claude Pro and Max via usage credits, Team and Enterprise
- #8Claude Opus 5Anthropic's previous-generation Opus model (July 2026), succeeded by Opus 5.5 in September 2026, with adaptive reasoning effort levels.Tier S, score 88.0API $5 input / $25 output per 1M tokens; listed as a legacy model in Claude plansAPI $5 input / $25 output per 1M tokens; listed as a legacy model in Claude plans
- #9Muse WonderNEWMeta's Muse reasoning model (October 2026) for complex software engineering and long-horizon agent work, offered through the Meta Model API.Tier S, score 87.7API about $4 input / $20 output per 1M tokens (Meta Model API preview, per Artificial Analysis)API about $4 input / $20 output per 1M tokens (Meta Model API preview, per Artificial Analysis)
- #10GPT-5.6 SolThe frontier-capability tier of OpenAI's GPT-5.6 family, aimed at coding agents and demanding professional work, with a 1.05M-token context window.Tier S, score 87.3API $4 input / $20 output per 1M tokens (promotional rate through at least Nov 21, 2026)API $4 input / $20 output per 1M tokens (promotional rate through at least Nov 21, 2026)
- #11Grok 4.7xAI's September 2026 model for coding, agentic tasks and knowledge work, with a 500K-token context window and four reasoning settings.Tier S, score 87.1API $2 input / $6 output per 1M tokens (prompts under 200K; $4 / $12 above)API $2 input / $6 output per 1M tokens (prompts under 200K; $4 / $12 above)
- #12Claude Fable 5Anthropic's June 2026 Fable model for demanding reasoning and long-horizon agentic work, now a legacy model succeeded by Fable 5.1 (September 2026).Tier S, score 87.0$10 / $50 per 1M tokens (in/out)$10 / $50 per 1M tokens (in/out)
- #13Qwen3.8 MaxAlibaba's 2.4T-parameter open-weight flagship placing top-5 on frontier leaderboards.Tier S, score 86.3Open weights (Qwen3.8-Max License); API $2 input / $6 output per 1M tokens via Alibaba Cloud Model Studio (international)Open weights (Qwen3.8-Max License); API $2 input / $6 output per 1M tokens via Alibaba Cloud Model Studio (international)
- #14GLM-5.3Z.ai's open-weight flagship with a 1M-token context window for long-horizon work.Tier S, score 86.3Open weights (GLM-5.3 License); API $1.40 input / $4.40 output per 1M tokens via Z.aiOpen weights (GLM-5.3 License); API $1.40 input / $4.40 output per 1M tokens via Z.ai
- #15GPT-6 SolOpenAI's first GPT-6 Sol release (September 22, 2026) for coding and agent workflows, followed a week later by GPT-6.1 Sol.Tier S, score 86.1API $2 input / $10 output per 1M tokens ($0.20 cached input)API $2 input / $10 output per 1M tokens ($0.20 cached input)
- #16Claude Haiku 5.5Anthropic's fastest Claude 5.5 model (October 7, 2026) for classification, extraction, routing and sub-agent work, with a 1M-token context window.Tier S, score 86.1API from $0.10 input / $0.50 output per 1M tokens (prompts up to 100K; $0.50 / $2.50 above)API from $0.10 input / $0.50 output per 1M tokens (prompts up to 100K; $0.50 / $2.50 above)
- #17DeepSeek V4.1 FlashDeepSeek's September 2026 Flash model on a new architecture, adding native image understanding and a 1M-token context at lower API prices.Tier S, score 85.7API $0.30 input / $1.20 output per 1M tokens at peak ($0.15 / $0.60 off-peak); MIT open weightsAPI $0.30 input / $1.20 output per 1M tokens at peak ($0.15 / $0.60 off-peak); MIT open weights
- #18Kimi K3Moonshot AI's 2.8T-parameter mixture-of-experts model and the highest-scoring open-weight entry on the Artificial Analysis Intelligence Index.Tier A, score 85.0Open weights on Hugging Face; API $3 input / $15 output per 1M tokens ($0.30 cached input)Open weights on Hugging Face; API $3 input / $15 output per 1M tokens ($0.30 cached input)
- #19GPT-5.6 TerraBalanced everyday member of the GPT-5.6 family, tuned for cost-aware agents.Tier A, score 84.8API $2 input / $12 output per 1M tokens ($0.20 cached input)API $2 input / $12 output per 1M tokens ($0.20 cached input)
- #20Claude Sonnet 5Anthropic's June 2026 Sonnet model for large-scale coding and agentic work, now a legacy model succeeded by Sonnet 5.5 (September 2026) at the same price.Tier A, score 84.8API $2 input / $10 output per 1M tokens (legacy model)API $2 input / $10 output per 1M tokens (legacy model)
Show all 81 modelsShow fewer
- #21Claude Opus 4.8An earlier Anthropic Opus release (May 2026), strong across coding, reasoning and long-context knowledge work; since followed by Opus 5 and 5.5.Tier A, score 84.7$5 / $25 per 1M tokens (in/out)$5 / $25 per 1M tokens (in/out)
- #22GPT-6 LunaOpenAI's low-cost GPT-6 tier for high-volume tasks such as extraction, summarization and routing, with a 1.05M-token context window.Tier A, score 84.1API $0.10 input / $0.50 output per 1M tokens ($0.01 cached input)API $0.10 input / $0.50 output per 1M tokens ($0.01 cached input)
- #23GPT-5.5OpenAI's April 2026 flagship for complex professional and agentic coding work, since followed by the GPT-5.6 family (July 2026) and GPT-6 Astra (September 2026).Tier A, score 84.0API $5 input / $30 output per 1M tokensAPI $5 input / $30 output per 1M tokens
- #24Grok 4.5xAI's 2026 frontier release pairing native tool use with real-time search.Tier A, score 83.8API $2 input / $6 output per 1M tokens (prompts under 200K; $4 / $12 above); also in Grok subscriptionsAPI $2 input / $6 output per 1M tokens (prompts under 200K; $4 / $12 above); also in Grok subscriptions
- #25MiMo-V2.6-ProNEWXiaomi's open-weight omnimodal mixture-of-experts model (September 2026) with a 1M-token context window, released under the MIT license.Tier A, score 83.8Open weights (MIT); API about $0.43 input / $0.87 output per 1M tokens (per Artificial Analysis)Open weights (MIT); API about $0.43 input / $0.87 output per 1M tokens (per Artificial Analysis)
- #26Gemini 3.1 Deep ThinkGoogle's deep-reasoning Gemini variant for science, research, and engineering.Tier A, score 83.6Included in the Google AI Ultra plan (from $99.99/mo); not on the Gemini API price listIncluded in the Google AI Ultra plan (from $99.99/mo); not on the Gemini API price list
- #27Mistral Medium 3.5Frontier-class multimodal model from Mistral optimized for agentic and coding use cases.Tier A, score 83.6Open weights (Modified MIT); API $1.50 input / $7.50 output per 1M tokens via MistralOpen weights (Modified MIT); API $1.50 input / $7.50 output per 1M tokens via Mistral
- #28Grok 4.6xAI's cost-efficient frontier model, posting frontier-level composite index results at $2/$6 per million tokens.Tier A, score 83.2$2 / $6 per 1M tokens (in/out)$2 / $6 per 1M tokens (in/out)
- #29Gemini 3.8 FlashGoogle's fast Flash model tuned for complex agentic tasks at scale.Tier A, score 83.2API $0.75 input / $3.75 output per 1M tokens through Dec 31, 2026 ($1.50 / $7.50 from Jan 1, 2027)API $0.75 input / $3.75 output per 1M tokens through Dec 31, 2026 ($1.50 / $7.50 from Jan 1, 2027)
- #30DeepSeek V4 ProMIT-licensed open-weight heavyweight, among the open-weight leaders on SWE-bench Verified and LiveCodeBench.Tier A, score 83.0API $1.32 input / $3.96 output per 1M tokens at peak hours ($0.66 / $1.98 off-peak); MIT open weightsAPI $1.32 input / $3.96 output per 1M tokens at peak hours ($0.66 / $1.98 off-peak); MIT open weights
- #31GPT-5.3 CodexOpenAI's February 2026 agentic coding model combining code generation with general-purpose intelligence, since folded into GPT-5.4 and followed by the GPT-5.5 and GPT-5.6 families.Tier A, score 82.9API $1.75 input / $14 output per 1M tokensAPI $1.75 input / $14 output per 1M tokens
- #32MiniMax M3.1 FlashNEWMiniMax's fast, low-cost model (October 19, 2026) for agentic tasks, coding and computer use, with a 1M-token context window.Tier A, score 82.5API $0.15 input / $0.60 output per 1M tokens ($0.015 cached input)API $0.15 input / $0.60 output per 1M tokens ($0.015 cached input)
- #33Claude Mythos PreviewAnthropic's invite-only cybersecurity research preview offered through Project Glasswing, deprecated on the Claude API since June 2026.Tier A, score 82.4Invite-only via Project Glasswing; $25 input / $125 output per 1M tokens after creditsInvite-only via Project Glasswing; $25 input / $125 output per 1M tokens after credits
- #34GPT-5.4 ProA model focused on pure reasoning and mathematical problem-solving, setting a record on the FrontierMath benchmark.Tier A, score 82.3API $5 / 1M tokens (input), $25 / 1M tokens (output)API $5 / 1M tokens (input), $25 / 1M tokens (output)
- #35Claude Opus 4.7Anthropic's April 2026 Opus release for agentic coding and long-context reasoning, now a legacy model succeeded by Opus 4.8, Opus 5 and Opus 5.5.Tier A, score 82.0API $5 input / $25 output per 1M tokens (legacy model)API $5 input / $25 output per 1M tokens (legacy model)
- #36Mistral Large 4Mistral's 1-trillion-parameter multimodal mixture-of-experts model (October 2026), combining instruct and reasoning modes, with open weights planned for late October.Tier A, score 81.8API $1.36 input / $4.18 output per 1M tokens (preview pricing)API $1.36 input / $4.18 output per 1M tokens (preview pricing)
- #37Nemotron 3 SuperNVIDIA's open-weight reasoning model available with first-party token pricing.Tier A, score 81.8Open weights (NVIDIA Nemotron Open Model License); API $0.08 input / $0.45 output per 1M tokens via OpenRouter (a free variant is also listed)Open weights (NVIDIA Nemotron Open Model License); API $0.08 input / $0.45 output per 1M tokens via OpenRouter (a free variant is also listed)
- #38Qwen 3.7 MaxAlibaba's Qwen 3.7 Max, a proprietary API model with a 1M-token context and thinking mode, priced below Claude Opus 5.5 and GPT-5.6 Sol per token.Tier A, score 81.7API $2.50 input / $7.50 output per 1M tokens (Alibaba Cloud Model Studio, international)API $2.50 input / $7.50 output per 1M tokens (Alibaba Cloud Model Studio, international)
- #39GPT-5.6 LunaOpenAI's cost-optimized GPT-5.6 tier, repriced 80% lower in July 2026 for high-volume production workloads.Tier A, score 81.4$0.20 per 1M input tokens, $1.20 per 1M output tokens$0.20 per 1M input tokens, $1.20 per 1M output tokens
- #40Claude Sonnet 4.6Anthropic's previous-generation Sonnet, since followed by Sonnet 5 and 5.5, balancing near-Opus quality with lower cost.Tier A, score 81.3Claude Pro $20/mo · API ~$3/$15 per 1M tokensClaude Pro $20/mo · API ~$3/$15 per 1M tokens
- #41Mistral Large 3Mistral's December 2025 open-weight (Apache 2.0) Mixture-of-Experts model with 675B total parameters and a 256K-token context window.Tier A, score 80.9API $0.50 input / $1.50 output per 1M tokens; open weights (Apache 2.0)API $0.50 input / $1.50 output per 1M tokens; open weights (Apache 2.0)
- #42MiniMax M3Natively multimodal open-weight model built for very long agent trajectories.Tier A, score 80.0Open weights (MiniMax Community License); API $0.30 input / $1.20 output per 1M tokens via MiniMax (prompts up to 512K; $0.60 / $2.40 above)Open weights (MiniMax Community License); API $0.30 input / $1.20 output per 1M tokens via MiniMax (prompts up to 512K; $0.60 / $2.40 above)
- #43Mistral Small 4Hybrid open model unifying instruct, reasoning, and coding in a single efficient package.Tier A, score 80.0Free (open weights); paid API availableFree (open weights); paid API available
- #44Gemini 3.6 FlashGoogle's fast-tier Gemini 3.6 model (ninth on the August 2026 BenchAlign leaderboard), since followed by Gemini 3.7 and 3.8 Flash.Tier A, score 79.9Free tier in the Gemini app; usage-based API pricingFree tier in the Gemini app; usage-based API pricing
- #45GLM-5.2Open-weight generalist tuned for long-horizon planning and coding work.Tier A, score 79.9Open weights (MIT); API $1.40 input / $4.40 output per 1M tokens via Z.aiOpen weights (MIT); API $1.40 input / $4.40 output per 1M tokens via Z.ai
- #46GPT-5.4OpenAI's March 2026 flagship for professional work, coding and computer use, since succeeded by GPT-5.5 and the GPT-5.6 family.Tier A, score 79.6API $2.50 input / $15 output per 1M tokensAPI $2.50 input / $15 output per 1M tokens
- #47MiniMax M2.7MiniMax's March 2026 open-weight agentic coding model, since succeeded by MiniMax M3 (June 2026).Tier A, score 79.2$0.30 / 1M tokens$0.30 / 1M tokens
- #48Gemini 3.1 ProA frontier model with strong reasoning performance, including high marks on the ARC-AGI-2 benchmark.Tier A, score 79.0API $2.00 / 1M tokens (input), $12.00 / 1M tokens (output)API $2.00 / 1M tokens (input), $12.00 / 1M tokens (output)
- #49GLM-5.1An open-weight model from Zhipu AI that offers performance competitive with leading closed models.Tier A, score 78.2$1.40 / 1M input, $4.40 / 1M output tokens$1.40 / 1M input, $4.40 / 1M output tokens
- #50Qwen3.5Alibaba's open-weight generalist model combining reasoning and non-reasoning modes with native vision input.Tier A, score 77.7Open weights (free to self-host) · cloud API usage-basedOpen weights (free to self-host) · cloud API usage-based
- #51GPT-5OpenAI's unified flagship model offered across all ChatGPT tiers with broad general-purpose capability.Tier A, score 77.4Free tier · ChatGPT Plus $20/mo · API ~$1.25/$10 per 1M tokensFree tier · ChatGPT Plus $20/mo · API ~$1.25/$10 per 1M tokens
- #52GPT-5.2OpenAI's December 2025 flagship for complex professional work, since followed by GPT-5.4, GPT-5.5, the GPT-5.6 family and GPT-6 Astra.Tier A, score 77.1API $1.75 input / $14 output per 1M tokensAPI $1.75 input / $14 output per 1M tokens
- #53Kimi K2.7 CodeCoding-focused open-weight mixture-of-experts model built for long agent runs.Tier A, score 76.9Open weights (Modified MIT); API $0.95 input / $4 output per 1M tokens via Moonshot ($0.19 cached input)Open weights (Modified MIT); API $0.95 input / $4 output per 1M tokens via Moonshot ($0.19 cached input)
- #54Nemotron 3 UltraNVIDIA's flagship open-weight model, positioned around deployment efficiency.Tier A, score 76.1Open weights; low-cost hosted routes incl. a free tierOpen weights; low-cost hosted routes incl. a free tier
- #55Kimi K2.6Trillion-parameter open-weight coding and agentic specialist from Moonshot AI, competitive with closed frontier models.Tier A, score 75.8Open weights (Modified MIT); API $0.95 input / $4 output per 1M tokens via Moonshot ($0.16 cached input)Open weights (Modified MIT); API $0.95 input / $4 output per 1M tokens via Moonshot ($0.16 cached input)
- #56Grok 4.20A model with native tool use and real-time search integration that is designed for rapid learning and weekly improvement.Tier A, score 75.4API $1.25 input / $2.50 output per 1M tokens (prompts under 200K; $2.50 / $5 above); also via Grok app subscriptionsAPI $1.25 input / $2.50 output per 1M tokens (prompts under 200K; $2.50 / $5 above); also via Grok app subscriptions
- #57Grok 4 HeavyxAI's July 2025 Grok 4 variant that runs parallel test-time compute, offered through the SuperGrok Heavy plan; since followed by Grok 4.20, 4.5, 4.6 and 4.7.Tier A, score 74.9SuperGrok Heavy subscription ($300/mo); not listed in the xAI API model catalogSuperGrok Heavy subscription ($300/mo); not listed in the xAI API model catalog
- #58GLM-5.3-FlashZ.ai's lighter open-weight GLM-5.3 variant with a 1M-token context window.Tier A, score 74.8Open weights (MIT); API $0.15 input / $0.50 output per 1M tokens via Z.aiOpen weights (MIT); API $0.15 input / $0.50 output per 1M tokens via Z.ai
- #59Llama 4The fourth generation of Meta's open-weight model family, a prominent choice in the 2026 open-source landscape.Tier A, score 74.8Open weights (Llama 4 Community License); API from $0.10 / $0.30 (Scout) to $0.19 / $0.65 (Maverick) per 1M tokens via OpenRouterOpen weights (Llama 4 Community License); API from $0.10 / $0.30 (Scout) to $0.19 / $0.65 (Maverick) per 1M tokens via OpenRouter
- #60GPT-5.1OpenAI's warmer, adaptive-reasoning update to GPT-5 with a 400K-token context window for everyday and agentic use.Tier A, score 74.4$1.25 / $10 per 1M tokens (input/output)$1.25 / $10 per 1M tokens (input/output)
- #61Gemini 2.5 ProGoogle's 2025 Pro model with a 1M-token context window, since followed by the Gemini 3 series; Google lists Gemini 3.1 Pro as its replacement.Tier A, score 74.3API $1.25 input / $10 output per 1M tokens (prompts up to 200K; $2.50 / $15 above)API $1.25 input / $10 output per 1M tokens (prompts up to 200K; $2.50 / $15 above)
- #62Grok BuildNEWxAI's coding model for agentic software engineering and workflow tasks, with a 256K-token context window and low API pricing.Tier A, score 73.8API $1 input / $2 output per 1M tokens (prompts under 200K; $2 / $4 above)API $1 input / $2 output per 1M tokens (prompts under 200K; $2 / $4 above)
- #63Claude Haiku 4.5Anthropic's lightweight Haiku model with a 200K-token context window.Tier A, score 73.4API $2.00 per 1M tokens blendedAPI $2.00 per 1M tokens blended
- #64Gemma 4 31BCompact open-weight model from Google with solid function calling for lightweight deployments.Tier A, score 73.0Free (open weights)Free (open weights)
- #65OpenAI o3OpenAI's April 2025 reasoning model for math, science and coding, deprecated in the API since June 2026 with shutdown set for December 2026.Tier A, score 72.8API $2 input / $8 output per 1M tokens (deprecated; shutdown December 11, 2026)API $2 input / $8 output per 1M tokens (deprecated; shutdown December 11, 2026)
- #66Qwen3-CoderOpen-weight coding specialist from Alibaba with 256K context built for agentic coding workflows.Tier A, score 71.7Free (open weights)Free (open weights)
- #67DeepSeek V4 FlashCost-efficient open-weight sibling of V4 aimed at production agent pipelines.Tier A, score 71.5Open weights (MIT); the deepseek-v4-flash API name now routes to V4.1-Flash at $0.30 input / $1.20 output per 1M tokens peak ($0.15 / $0.60 off-peak)Open weights (MIT); the deepseek-v4-flash API name now routes to V4.1-Flash at $0.30 input / $1.20 output per 1M tokens peak ($0.15 / $0.60 off-peak)
- #68GPT-5.4 miniFaster smaller-tier reasoning with wide availabilityTier A, score 71.3Free tier · Plus $20/moFree tier · Plus $20/mo
- #69Gemini 3 FlashGoogle's original Gemini 3 Flash model, since followed by Gemini 3.5 through 3.8 Flash.Tier A, score 71.1Free with Google accountFree with Google account
- #70gpt-oss-120bOpenAI's Apache 2.0 open-weight reasoning model (August 2025) that fits on a single 80GB GPU, with configurable reasoning effort.Tier A, score 70.5Free open weights (Apache 2.0); hosted API pricing varies by providerFree open weights (Apache 2.0); hosted API pricing varies by provider
- #71Mistral LargeOpen-weight Mistral model competitive on business writing tasksTier B, score 69.2Free · Pro from $14.99/moFree · Pro from $14.99/mo
- #72Qwen3.5-397B-A17BA large, open-source model from Alibaba Cloud with competitive performance on leaderboards.Tier B, score 68.5$0.60 / 1M input, $3.60 / 1M output tokens$0.60 / 1M input, $3.60 / 1M output tokens
- #73Command A+Cohere's first Mixture-of-Experts model, combining vision input, agentic tool use, reasoning and translation in one enterprise deployment.Tier B, score 68.3Open weights (Apache 2.0); API $0.30 input / $1.50 output per 1M tokens via OpenRouter; also on Azure AI FoundryOpen weights (Apache 2.0); API $0.30 input / $1.50 output per 1M tokens via OpenRouter; also on Azure AI Foundry
- #74Ministral 3 14BCompact open-weight text and vision model for efficient edge deployment.Tier B, score 68.3Open weights, free to self-hostOpen weights, free to self-host
- #75Gemini 2.5 FlashGoogle's fast, low-cost multimodal Gemini variant with a 1M-token context window.Tier B, score 68.2Free in Gemini app · API ~$0.30/$2.50 per 1M tokensFree in Gemini app · API ~$0.30/$2.50 per 1M tokens
- #76Llama 4 MaverickCost-sensitive open-weight Llama 4 variant aimed at multimodal general-purpose deployments.Tier B, score 65.7Open weights (Llama 4 Community License); API $0.19 input / $0.65 output per 1M tokens via OpenRouterOpen weights (Llama 4 Community License); API $0.19 input / $0.65 output per 1M tokens via OpenRouter
- #77GPT-5 miniOpenAI's cost-efficient GPT-5 variant aimed at general-purpose tasks with a 400K-token context.Tier B, score 64.9Bundled with ChatGPT plans · API ~$0.25/$2.00 per 1M tokensBundled with ChatGPT plans · API ~$0.25/$2.00 per 1M tokens
- #78Ministral 3 8BEfficient open-weight text and vision model for lightweight workloads.Tier B, score 64.3Open weights, free to self-hostOpen weights, free to self-host
- #79GPT-4oOpenAI's 2024 multimodal GPT model, retired from ChatGPT in early 2026 and still available in the API.Tier B, score 60.4API $2.50 input / $10 output per 1M tokens; no longer offered in ChatGPTAPI $2.50 input / $10 output per 1M tokens; no longer offered in ChatGPT
- #80Llama 4 ScoutAn open-source model designed for massive-scale data processing with a 10 million token context window.Tier B, score 59.5Open Source (Free to use, self-hosted)Open Source (Free to use, self-hosted)
- #81Cohere Command R+Cohere's 2024 RAG-tuned enterprise model; Cohere now points new projects to its newer Command A models, led by Command A+.Tier C, score 54.8Free trial key (non-commercial); API $2.50 input / $10 output per 1M tokens (08-2024 version)Free trial key (non-commercial); API $2.50 input / $10 output per 1M tokens (08-2024 version)