{"umans-coder":{"name":"umans-coder","display_name":"Umans Coder","description":"Our recommended model for complex, coding-heavy workloads. Optimized for the best experience with coding agents like Claude Code.","base_model":{"name":"DeepSeek-V4.1-Flash-1017","provider":"DeepSeek","oss_base":"DeepSeek-V4.1-Flash-1017"},"capabilities":{"max_completion_tokens":393216,"recommended_max_tokens":393215,"context_window":1048576,"supports_vision":true,"supports_tools":true,"reasoning":{"supported":true,"can_disable":true,"levels":["none","low","high","max"],"default_level":"high"}},"benchmarks":{},"weights":{"precision":"fp8","hf_url":"https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"}},"umans-flash":{"name":"umans-flash","display_name":"Umans Flash","description":"Our fast tier for the roles around your main coder: scout subagents, context gathering, research, summaries, documentation and quick edits. It runs the same model as umans-coder today, DeepSeek V4.1 Flash, which is fast enough to fill both roles.","base_model":{"name":"DeepSeek-V4.1-Flash-1017","provider":"DeepSeek","oss_base":"DeepSeek-V4.1-Flash-1017"},"capabilities":{"max_completion_tokens":262144,"recommended_max_tokens":32768,"context_window":262144,"supports_vision":true,"supports_tools":true,"reasoning":{"supported":true,"can_disable":true,"levels":["none","low","high","max"],"default_level":"high"}},"benchmarks":{},"weights":{"precision":"fp8","hf_url":"https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"}},"umans-deepseek-v4.1-flash":{"name":"umans-deepseek-v4.1-flash","display_name":"Umans DeepSeek V4.1 Flash","lifecycle":{"production_start_date":"2026-09-18","updates":[{"date":"2026-10-20","label":"1017","weights":{"precision":"fp8","hf_url":"https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"},"lab":"umans-deepseek-v4.1-flash-1017-lab"}]},"description":"DeepSeek V4.1 Flash, served from DeepSeek's official DeepSeek-V4.1-Flash-1017 release: the October 17 update of DeepSeek's latest flash model, a 552B mixture-of-experts (8B active per token on prefill, 16B on decode) built for fast agentic coding, with native image understanding and a 1M-token context window. Updated on October 20, 2026 from the original DeepSeek-V4.1-Flash weights to 1017, which DeepSeek reports as stronger on reasoning, coding, agentic and visual benchmarks, with shorter outputs. This id follows future V4.1 Flash updates; umans-deepseek-v4.1-flash-1017 stays on 1017. Reasoning has four modes: non-think (none), think low (low), think high (high, the default) and think max (max). Billed per token ($0.15 / $0.60 / $0.028 per 1M; input / output / cache read). Served on our own GPU infrastructure with high availability.","base_model":{"name":"DeepSeek-V4.1-Flash-1017","provider":"DeepSeek","oss_base":"DeepSeek-V4.1-Flash-1017"},"capabilities":{"max_completion_tokens":393216,"recommended_max_tokens":393215,"context_window":1048576,"supports_vision":true,"supports_tools":true,"reasoning":{"supported":true,"can_disable":true,"levels":["none","low","high","max"],"default_level":"high"}},"benchmarks":{},"weights":{"precision":"fp8","hf_url":"https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"}},"umans-deepseek-v4.1-flash-1017":{"name":"umans-deepseek-v4.1-flash-1017","display_name":"Umans DeepSeek V4.1 Flash 1017","lifecycle":{"production_start_date":"2026-10-20"},"description":"DeepSeek V4.1 Flash pinned to its 1017 update, DeepSeek's official DeepSeek-V4.1-Flash-1017 release: the same model umans-deepseek-v4.1-flash serves since October 20, 2026, under an id that stays on this checkpoint when V4.1 Flash moves to a later update. A 552B mixture-of-experts (8B active per token on prefill, 16B on decode) built for fast agentic coding, with native image understanding and a 1M-token context window. Reasoning has four modes: non-think (none), think low (low), think high (high, the default) and think max (max). Billed per token ($0.15 / $0.60 / $0.028 per 1M; input / output / cache read). Served on our own GPU infrastructure with high availability.","base_model":{"name":"DeepSeek-V4.1-Flash-1017","provider":"DeepSeek","oss_base":"DeepSeek-V4.1-Flash-1017"},"capabilities":{"max_completion_tokens":393216,"recommended_max_tokens":393215,"context_window":1048576,"supports_vision":true,"supports_tools":true,"reasoning":{"supported":true,"can_disable":true,"levels":["none","low","high","max"],"default_level":"high"}},"benchmarks":{},"weights":{"precision":"fp8","hf_url":"https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"}},"umans-glm-5.3":{"name":"umans-glm-5.3","display_name":"Umans GLM 5.3","lifecycle":{"production_start_date":"2026-08-29"},"description":"GLM 5.3 is Z.ai's flagship open-weights model for coding and long-horizon agentic work: a 744B mixture-of-experts with 40B active parameters and a 1M-token context window, the successor to GLM 5.2 on the same base with large post-training gains on complex coding tasks. It always thinks at max effort by default; dial reasoning to low or high (thinking cannot be turned off).","base_model":{"name":"GLM-5.3","provider":"Z.ai","oss_base":"GLM-5.3"},"capabilities":{"max_completion_tokens":131072,"recommended_max_tokens":131071,"context_window":1048576,"supports_vision":false,"supports_tools":true,"reasoning":{"supported":true,"can_disable":false,"levels":["low","high","max"],"default_level":"max"}},"benchmarks":{},"weights":{"precision":"fp8","hf_url":"https://huggingface.co/zai-org/GLM-5.3"}},"umans-kimi-k3.1-pro-lab":{"name":"umans-kimi-k3.1-pro-lab","display_name":"Umans Kimi K3.1 Pro (lab)","stage":"playground","description":"Kimi K3.1 Pro as a Labs experiment, open for a short test window: temporary, not a permanent id. Moonshot's newest open model, post-trained from Kimi K3 and served from the official Kimi-K3.1-Pro open weights: a 2.8T mixture-of-experts (104B active per token) with native image understanding and a 1M-token context, which Moonshot reports as substantially stronger than K3 on coding, factual accuracy and agentic work. It thinks by default at maximum reasoning effort; select none, low, high or max to trade depth for speed. Access is seat-gated through the Labs page while an experiment is live. It is offered at limited capacity and low availability, so expect it to be flaky and to go down under load: crash it, give it a moment, and try again. For production work we recommend the pay-per-token umans-kimi-k3.1-pro.","base_model":{"name":"Kimi-K3.1-Pro","provider":"Moonshot","oss_base":"Kimi K3.1 Pro"},"capabilities":{"max_completion_tokens":131072,"recommended_max_tokens":131071,"context_window":1048576,"supports_vision":true,"supports_tools":true,"reasoning":{"supported":true,"can_disable":true,"levels":["none","low","high","max"],"default_level":"max"}},"benchmarks":{},"weights":{"precision":"full","hf_url":"https://huggingface.co/moonshotai/Kimi-K3.1-Pro"}},"umans-kimi-k3.1-pro":{"name":"umans-kimi-k3.1-pro","display_name":"Umans Kimi K3.1 Pro","lifecycle":{"production_start_date":"2026-11-06"},"description":"Kimi K3.1 Pro: Moonshot's newest open model and the first of the K3.1 series, post-trained from Kimi K3 and served from the official Kimi-K3.1-Pro open weights. A 2.8T mixture-of-experts (104B active per token) with native image understanding and a 1M-token context window, which Moonshot reports as substantially stronger than K3 on coding, factual accuracy and agentic work. It thinks by default at maximum reasoning effort; select none, low, high or max to trade depth for speed.","base_model":{"name":"Kimi-K3.1-Pro","provider":"Moonshot","oss_base":"Kimi K3.1 Pro"},"capabilities":{"max_completion_tokens":131072,"recommended_max_tokens":131071,"context_window":1048576,"supports_vision":true,"supports_tools":true,"reasoning":{"supported":true,"can_disable":true,"levels":["none","low","high","max"],"default_level":"max"}},"benchmarks":{},"weights":{"precision":"full","hf_url":"https://huggingface.co/moonshotai/Kimi-K3.1-Pro"}}}