AI Findings · Data Digest
Findings from BenchGecko's own tests and price tracking as they happen, and a weekly digest of new models, price changes and benchmark results. Every number links to its data.
Beam 1.0 added to BenchGecko · Reflection 501B open-weight MoE with 23B active and 1M context
Reflection released Beam 1.0 open weights on 2026-10-16 under Apache 2.0. BenchGecko lists it at $0.25 input and $1.00 output per M tokens with a 1,048,576 token context window.
Kimi K3 price cut 53%
Kimi K3's list price dropped 53% (input $0.95/M to $0.45/M, output $14/M to $15/M), confirmed in the next daily snapshot.
Jais 3 added to BenchGecko · six open Arabic-centric models from 0.9B to 375B
MBZUAI Institute of Foundation Models released Jais 3 on 2026-10-15: six open-weight Arabic-centric models under Apache 2.0, up to 375B total and 23B active parameters, with 512K context on five of the six.
Muse Spark Open added to BenchGecko · Meta 816B open-weight MoE with 42B active and 1M context
Meta released open weights for Muse Spark Open on 2026-10-15. BenchGecko lists it at $1.25 input and $4.25 output per M tokens with a 1,048,576 token context window.
Step 5 Preview weights now open · StepFun 600B MoE with 27B active and 1M context
StepFun published open weights for Step 5 Preview on 2026-10-15. BenchGecko lists it at $1 input and $2.70 output per M tokens with a 1,048,576 token context window.
Nemotron 3.5 Super VL and Bolt 74B-A9B added to BenchGecko · NVIDIA open multimodal MoE models
NVIDIA released Nemotron 3.5 Super VL (120B total, 12B active, up to 1M context) and Nemotron 3.5 Bolt 74B-A9B (24GB, 256K context) as open weights on 2026-10-15.
Ultramarine 1 Flash added to BenchGecko · knowledge-work model at $0.15/$0.50 per M
Paper Instruments launched Ultramarine 1 Flash on 2026-10-14, a GLM-5.3 Flash post-train for professional work. BenchGecko lists it at $0.15/$0.50 per M tokens.
LFM2.5-24B-A2B added to BenchGecko · Liquid AI reasoning MoE, 2B active
Liquid AI released LFM2.5-24B-A2B on 2026-10-13. BenchGecko lists it at $0.05/$0.70 per M tokens with a 128,000 token context window.
MiniCPM-V 4.7 35B-A3B added to BenchGecko · open-weight multimodal MoE
OpenBMB released MiniCPM-V 4.7 on 2026-10-12. BenchGecko lists the 35B-A3B variant, an Apache-2.0 multimodal MoE with a 262,144 token position limit.
Muse Wonder added to BenchGecko · Meta frontier model at $4/$20 per M
Meta launched Muse Wonder on 2026-10-12 on Meta Model API and Muse Code. BenchGecko lists it at $4 input, $0.20 cached input and $20 output per M tokens.
Gemini 4 Argon added to BenchGecko · 1,048,576 context, $2/$10 per M introductory
Google made Gemini 4 Argon generally available on 2026-10-13. BenchGecko lists it at $2/$10 per M tokens (introductory) with a 1,048,576 token context window.
Mistral Large 4 added to BenchGecko · specs, launch pricing, vendor benches
Mistral AI opened a public preview of Mistral Large 4 on 2026-10-06. BenchGecko lists the model at $0.68/$2.09 per M tokens (sale) with a 1,048,576 token context.
Microsoft Decision 1 added to BenchGecko · decision scoring at $0.042 per M input
Microsoft released Microsoft-Decision-1 on 2026-10-09 for structured decision scoring. BenchGecko lists it at $0.042 per M input tokens, free output, and 32,768 context.
Claude Haiku 5.5 launch report card · BenchGecko's own tests on day one
Anthropic's Claude Haiku 5.5 went through BenchGecko's profile tests within hours of release: identity, refusals, knowledge cutoff and language cost. Price: $0.09999999999999999/M input and $0.5/M output (list price).
DeepSeek V4 Flash 0731 identifies as another lab's model in 2 of 6 answers
Asked what it is and who made it, DeepSeek's DeepSeek V4 Flash 0731 named Google and Anthropic instead of DeepSeek in 2 of 6 replies (run 2026-10-04, no system prompt).
DeepSeek V4.1 Flash identifies as another lab's model in 2 of 5 answers
Asked what it is and who made it, DeepSeek's DeepSeek V4.1 Flash named OpenAI instead of DeepSeek in 2 of 5 replies (run 2026-10-04, no system prompt).
Llama 4 Maverick identifies as another lab's model in 3 of 6 answers
Asked what it is and who made it, Meta's Llama 4 Maverick named Google and OpenAI instead of Meta in 3 of 6 replies (run 2026-10-04, no system prompt).
DeepSeek V4 Pro 0813 identifies as another lab's model in 3 of 6 answers
Asked what it is and who made it, DeepSeek's DeepSeek V4 Pro 0813 named OpenAI and Anthropic instead of DeepSeek in 3 of 6 replies (run 2026-10-04, no system prompt).
This week in AI · 2026-09-26 to 2026-10-02
Six new models were listed, from Anthropic and OpenAI. BenchGecko recorded 26 price changes, 12 of them drops, and added benchmark results for 198 models.