Value
MiniMax M3.1 Flash MiniMax-M3.1-Flash is a multimodal model from MiniMax.
1M context Free on OpenRouter Text, Image, Video → Text External rankings
Arena Web & Games #33 1587 score
What changed
#33 on the Arena Web & Games leaderboard From the announcement
MiniMax: flat API price $0.15 in / $0.60 out per 1M tokens; agentic-task cost 70% lower than its predecessor MiniMax: 73.7% on DeepSWE v1.1 and 61.7 on OSWorld V2; weights to be released later Value
DeepSeek V4.1 Flash 1017 Oct 17, 2026 · 3 days ago · Replaces DeepSeek V4.1 Flash (0910) DeepSeek V4.1 Flash 1017 is the October 17, 2026 snapshot of DeepSeek V4.1 Flash, served by DeepSeek under the `deepseek-flash` API model name.
1.05M context $0.15 in · $0.60 out Text, Image → Text External rankings
Arena Web & Games #13 1664 score What changed
#13 on the Arena Web & Games leaderboard Arena Web & Games score +46 vs DeepSeek V4.1 Flash (0910) (#13, was #26) From the announcement
DeepSeek reports stronger benchmark results, faster task completion and shorter outputs than the previous version Now serves the deepseek-flash API name; the retired V4-Flash names route to it DeepSeek API price: $0.15 in / $0.60 out per 1M tokens off-peak, double at peak hours Open weights on Hugging Face and ModelScope Open weights
Muse Spark Open Muse Spark Open is a coding-optimized open-weight model.
1.05M context $1.25 in · $4.25 out Text, Image, Files, Audio, Video → Text External rankings
Arena Web & Games #52 1528 score What changed
#52 on the Arena Web & Games leaderboard From the announcement
Meta: open weights for a natively multimodal model with 816B total parameters (42B active) and a 1M-token context window Meta: released under the Muse Model License with limited commercial terms; weights on Hugging Face and major clouds Flagship
Gemini 4 Argon Gemini 4 Argon is Google's flagship reasoning model, built for complex software engineering, deep mathematical and scientific problem solving, and autonomous multi-step agents.
1.05M context $2 in · $10 out Text, Image, Files, Audio, Video → Text External rankings
Arena Chat #1 1532 score Arena Coding Agents #6 Arena Web & Games #10 1678 score What changed
#1 on the Arena Chat leaderboard #6 on the Arena Coding Agents leaderboard From the announcement
Generally available Oct 13, 2026, after a Sep 30 rollout to trusted cyber defenders Google: output limit raised to 1M tokens, up from the previous 64K Flagship
Muse Wonder Muse Wonder is a frontier reasoning model from Meta, designed for complex personal and professional agentic tasks.
1.05M context $4 in · $20 out Text, Image, Files, Audio, Video → Text External rankings
Arena Chat #7 1501 score Arena Coding Agents #12 Arena Web & Games #17 1652 score What changed
#7 on the Arena Chat leaderboard #12 on the Arena Coding Agents leaderboard Arena Chat: down 2 since the Oct 13 update (#5 → #7) From the announcement
New top tier that builds on Meta's Muse Spark models Adds Active Memory and Active Focus; Muse Wonder Full is available to Muse Code subscribers Fast
Claude Haiku 5.5 Oct 7, 2026 · 13 days ago · Compared with Claude Haiku 4.5 Claude Haiku 5.5 is Anthropic's small, fast model for high-volume, cost-sensitive work such as summarization, subagents, and browser use.
1M context $0.10 in · $0.50 out Text, Image, Files → Text External rankings
Arena Coding Agents #25 Arena Web & Games #34 1586 score What changed
#25 on the Arena Coding Agents leaderboard Arena Web & Games score +256 vs Claude Haiku 4.5 (#34, was #123) Cheaper by 90%: $0.10 in / $0.50 out per 1M tokens (was $1 / $5 for Claude Haiku 4.5) Context 200K → 1M tokens From the announcement
Anthropic: costs around 75% less to run than Haiku 4.5 on average Anthropic: Terminal-Bench 4.0 39.2%, up from 0.0% for Haiku 4.5 Open weights
Mistral Large 4 Mistral Large 4 is a frontier multimodal (text and image input) model from Mistral AI built for reasoning, coding, and agentic workloads.
1.05M context $1.36 in · $4.18 out Text, Image → Text External rankings
Arena Coding Agents #39 Arena Vision #42 1272 score Arena Web & Games #45 1543 score Arena Chat #94 1430 score What changed
#39 on the Arena Coding Agents leaderboard From the announcement
Public preview on Mistral's API; Mistral says the open weights follow by the end of October Everyday
GPT-6.1 Sol Sep 29, 2026 · 21 days ago · Replaces GPT-6 Sol GPT-6.1 Sol is an upgrade to GPT-6 Sol from OpenAI, positioned below the flagship GPT-6 Astra in the GPT-6 series.
1.05M context $2 in · $10 out Text, Image, Files → Text External rankings
Arena Web & Games #4 1753 score Arena Coding Agents #4 Arena Vision #29 1288 score Arena Chat #57 1448 score Terminal-Bench 4.0 #3 58.18% in Codex What changed
#4 on the Arena Web & Games leaderboard #4 on the Arena Coding Agents leaderboard Arena Web & Games score +70 vs GPT-6 Sol (#4, was #9) Arena Vision score +48 vs GPT-6 Sol (#29, was #74) Same price as GPT-6 Sol: $2 in / $10 out per 1M tokens Terminal-Bench 4.0: 58.18% running in Codex (#3) From the announcement
OpenAI: nearly matches GPT-6 Astra on agentic coding, computer use and professional work at a fifth of Astra's token prices Cached input $0.10 per 1M tokens, 50% less than GPT-6 Sol Frontier
GPT-6.1 Sol Pro GPT-6.1 Sol Pro is the same underlying model as [GPT-6.1 Sol](https://openrouter.ai/openai/gpt-6.1-sol), served with `reasoning.mode` set to `pro` for higher-quality responses on complex tasks. **Cost note:** pro mode spends far more...
1.05M context $2 in · $10 out Text, Image, Files → Text From the announcement
OpenAI: GPT-6.1 Sol with reasoning.mode set to pro, for difficult tasks that need more model work and can tolerate higher latency and token usage OpenAI: GPT-6.1 Sol itself is an upgrade to GPT-6 Sol that nearly matches GPT-6 Astra on agentic coding, computer use and professional work Everyday
Claude Sonnet 5.5 Sep 28, 2026 · 22 days ago · Replaces Claude Sonnet 5 Claude Sonnet 5.5 is Anthropic's Sonnet-class model for well-scoped everyday work, succeeding Claude Sonnet 5 as a direct upgrade.
1M context $2 in · $10 out Text, Image, Files → Text External rankings
Arena Web & Games #3 1768 score Arena Coding Agents #5 Arena Chat #30 1470 score Arena Vision #30 1287 score Terminal-Bench 4.0 #2 61.82% in Claude Code What changed
#3 on the Arena Web & Games leaderboard #5 on the Arena Coding Agents leaderboard Arena Web & Games score +228 vs Claude Sonnet 5 (#3, was #47) Arena Chat score +28 vs Claude Sonnet 5 (#30, was #72) Arena Vision score +14 vs Claude Sonnet 5 (#30, was #41) Same price as Claude Sonnet 5: $2 in / $10 out per 1M tokens From the announcement
Anthropic: Terminal-Bench 4.0 70.6%, up from 10.3% for Sonnet 5 Runs 30%+ faster than Sonnet 5 and costs up to 30% less for most work Fast
Qwen3.8 Max Prime Qwen3.8 Max Prime is a higher-throughput variant of Qwen3.8 Max from Alibaba's Qwen team, served as a separate SKU at a higher price point.
1M context $4 in · $12 out Text, Image, Video → Text From the announcement
Alibaba Cloud: Prime mode of Qwen3.8-Max, with output speed raised to 1.5 to 2 times the standard API Alibaba Cloud: same capabilities and limits as the standard model, billed per input and output token Fast
GLM 5.3 Prime GLM-5.3-Prime is the high-speed variant of Z.ai's GLM-5.3, inheriting its full capabilities while delivering 1.5–2× the output throughput through inference acceleration.
1M context $2.80 in · $8.80 out Text → Text From the announcement
Alibaba Cloud Model Studio: Prime-mode GLM-5.3 with output speed 1.5 to 2 times the standard API Same capabilities and limits as GLM-5.3, which Z.ai describes as its flagship for complex software engineering and agent work Flagship
Claude Opus 5.5 Sep 22, 2026 · 28 days ago · Replaces Claude Opus 5 Claude Opus 5.5 is Anthropic's flagship model for demanding reasoning, coding, and long-horizon agentic work, succeeding Claude Opus 5.
1M context $4 in · $20 out Text, Image, Files → Text External rankings
Arena Web & Games #1 1807 score Arena Coding Agents #1 Arena Chat #2 1516 score Arena Vision #2 1321 score Terminal-Bench 4.0 #1 64.85% in Claude Code
What changed
#1 on the Arena Web & Games leaderboard #1 on the Arena Coding Agents leaderboard Arena Web & Games score +118 vs Claude Opus 5 (#1, was #8) Arena Chat score +8 vs Claude Opus 5 (#2, was #4) Cheaper by 20%: $4 in / $20 out per 1M tokens (was $5 / $25 for Claude Opus 5) Terminal-Bench 4.0: 64.85% running in Claude Code (#1) From the announcement
Anthropic: performs at the level of Claude Fable 5.1 on most work and costs about 40% less to run than Opus 5 Generates output more than 30% faster than Opus 5 Fast
GPT-6 Luna Sep 22, 2026 · 28 days ago · Replaces GPT-5.6 Luna GPT-6 Luna is the fast, cost-efficient model in OpenAI's GPT-6 series, positioned below GPT-6 Sol.
1.05M context $0.10 in · $0.50 out Text, Image, Files → Text External rankings
Arena Coding Agents #18 Arena Web & Games #38 1578 score Arena Vision #85 1216 score Arena Chat #166 1392 score Terminal-Bench 4.0 #31 16.36% in Codex What changed
#18 on the Arena Coding Agents leaderboard Arena Web & Games score +60 vs GPT-5.6 Luna (#38, was #56) Cheaper by 58%: $0.10 in / $0.50 out per 1M tokens (was $0.20 / $1.20 for GPT-5.6 Luna) Terminal-Bench 4.0: 16.36% running in Codex (#31) From the announcement
OpenAI: API price cut 50% versus GPT-5.6 Luna's promotional pricing, to $0.10 in / $0.50 out per 1M tokens OpenAI: on AutomationBench at high effort, 5.4 points better than its predecessor at 58% lower cost per task Flagship
Grok 4.7 Sep 21, 2026 · 29 days ago · Replaces Grok 4.6 Grok 4.7 is SpaceXAI's flagship model for coding, agentic tasks, and knowledge work, succeeding Grok 4.6.
500K context $2 in · $6 out Text, Image, Files → Text External rankings
Arena Coding Agents #16 Arena Web & Games #18 1637 score Arena Vision #79 1231 score Arena Chat #153 1401 score Terminal-Bench 4.0 #21 37.58% in Grok Build What changed
#16 on the Arena Coding Agents leaderboard #18 on the Arena Web & Games leaderboard Arena Web & Games score +20 vs Grok 4.6 (#18, was #27) Same price as Grok 4.6: $2 in / $6 out per 1M tokens Terminal-Bench 4.0: 37.58% running in Grok Build (#21) From the announcement
Served at the same price and speed as Grok 4.6 Flagship
GPT-6 Astra Sep 3, 2026 · 47 days ago · Compared with GPT-5.6 Sol GPT-6 Astra is OpenAI's flagship model for demanding end-to-end work.
1.05M context $10 in · $50 out Text, Image, Files → Text External rankings
Arena Web & Games #2 1774 score Arena Coding Agents #2 Arena Docs #16 1468 score Arena Vision #36 1280 score Arena Chat #69 1443 score Terminal-Bench 4.0 #3 58.18% in Codex What changed
#2 on the Arena Web & Games leaderboard #2 on the Arena Coding Agents leaderboard Arena Web & Games score +158 vs GPT-5.6 Sol (#2, was #28) Arena Docs score -15 vs GPT-5.6 Sol (#16, was #10) Pricier by 150%: $10 in / $50 out per 1M tokens (was $4 / $20 for GPT-5.6 Sol) Terminal-Bench 4.0: 58.18% running in Codex (#3) From the announcement
OpenAI: Terminal-Bench 4.0 57.9%, up from 37.3% for GPT-5.6 Sol OpenAI: OSWorld 2.0 72.6% in about 47% less time per task than GPT-5.6 Sol (65.7%) Flagship
Qwen3.8 Max (0902) Sep 2, 2026 · 48 days ago · Compared with Qwen3.7-Max Qwen3.8 Max 0902 is an updated snapshot of Qwen3.8 Max from Alibaba's Qwen team.
1M context $2 in · $6 out Text, Image, Video → Text External rankings
Arena Web & Games #11 1674 score Terminal-Bench 4.0 #25 26.97% in Claude Code What changed
#11 on the Arena Web & Games leaderboard Arena Web & Games score +159 vs Qwen3.7-Max (#11, was #57) Pricier by 36%: $2 in / $6 out per 1M tokens (was $1.48 / $4.42 for Qwen3.7-Max) New inputs: image, video Terminal-Bench 4.0: 26.97% running in Claude Code (#25) From the announcement
Sep 2 snapshot of Qwen3.8-Max: deeper coding, long-horizon autonomous development and stronger multi-tool agent work The qwen3.8-max endpoint moves to this snapshot with pricing unchanged Frontier
Claude Fable 5.1 Sep 1, 2026 · 49 days ago · Replaces Claude Fable 5 Claude Fable 5.1 improves on Claude Fable 5 across the board, with the biggest gains in agentic coding, long-running agentic workflows, and knowledge work: long code refactors, front-end and visual...
1M context $10 in · $50 out Text, Image, Files → Text External rankings
Arena Docs #2 1513 score Arena Chat #3 1512 score Arena Coding Agents #3 Arena Vision #4 1319 score Arena Web & Games #5 1743 score Terminal-Bench 4.0 #5 57.88% in Claude Code
What changed
#2 on the Arena Docs leaderboard #3 on the Arena Chat leaderboard Arena Docs score +17 vs Claude Fable 5 (#2, was #5) Arena Chat score +21 vs Claude Fable 5 (#3, was #10) Arena Vision score -6 vs Claude Fable 5 (#4, was #1) Arena Web & Games score +119 vs Claude Fable 5 (#5, was #23) From the announcement
Anthropic: an estimated 25% cheaper than Fable 5 for typical workloads, and up to about 45% on agentic work Open weights
GLM 5.3 Aug 14, 2026 · 67 days ago · Replaces GLM 5.2 GLM-5.3 is a large-scale reasoning model from Z.ai, built for complex software engineering and long-horizon agent tasks.
1.05M context $0.079 in · $1.40 out Text → Text External rankings
Arena Web & Games #25 1621 score Arena Coding Agents #26 Arena Chat #27 1471 score Terminal-Bench 4.0 #20 41.82% in Claude Code What changed
#25 on the Arena Web & Games leaderboard Arena Web & Games score +17 vs GLM 5.2 (#25, was #30) Cheaper by 42%: $0.079 in / $1.40 out per 1M tokens (was $0.62 / $2.42 for GLM 5.2) Terminal-Bench 4.0: 41.82% running in Claude Code (#20) From the announcement
Z.ai: same base model as GLM-5.2, with every gain from post-training Z.ai: 50% better than GLM-5.2 on its in-house Z.ai Code Bench Open weights
Kimi K3 Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI.
1.05M context $0.239 in · $11.94 out Text, Image, Video → Text External rankings
Arena Web & Games #16 1655 score Arena Coding Agents #23 Arena Chat #25 1475 score What changed
#16 on the Arena Web & Games leaderboard From the announcement
Full open weights released by July 27, 2026