Thirdshift

Limitless AI in an AI-less world.

A community network for affordable open-model access outside the AI bubble — for builders in places big labs ignore. Powered by idle gaming PCs. Alpha is reviewed by hand.

Machines online
Regions
Completions (24h)
Tokens served (24h)
Your region

Models

Growing catalog while host capacity comes online. Apply to reserve access — hosts stay private.

Llama 3.1 8B Instruct

Meta's everyday open chat model — strong general assistant for light workloads.

8,192 context

Tech specs

llama-3.1-8b-instruct · 2,048 max output

Typical hosted: $0.05 / $0.12

Built with Llama · Llama 3.1 is licensed under the Llama Community License · Acceptable Use Policy

$0.03 in / $0.08 out

per 1M tokens

~35% cheaper

Available on request

Expected 35 tok/s

Llama 3.3 70B Instruct

Frontier-class open reasoning and writing when smaller models fall short.

8,192 context

Tech specs

llama-3.3-70b-instruct · 2,048 max output

Typical hosted: $0.50 / $1.20

Built with Llama · Llama 3.3 is licensed under the Llama Community License · Acceptable Use Policy

$0.35 in / $0.90 out

per 1M tokens

~30% cheaper

Available on request

Expected 12 tok/s

Qwen2.5 72B Instruct

Alibaba's large open generalist — long-form reasoning, multilingual, and tool-friendly.

32,768 context

Tech specs

qwen2.5-72b-instruct · 4,096 max output

Typical hosted: $0.50 / $1.20

$0.35 in / $0.90 out

per 1M tokens

~30% cheaper

Available on request

Expected 12 tok/s

Qwen2.5 Coder 32B Instruct

Heavier code model for refactors, multi-file reasoning, and tougher programming tasks.

32,768 context

Tech specs

qwen2.5-coder-32b-instruct · 4,096 max output

Typical hosted: $0.22 / $0.65

$0.15 in / $0.45 out

per 1M tokens

~30% cheaper

Available on request

Expected 18 tok/s

DeepSeek R1 Distill 32B

Reasoning-first distill of DeepSeek R1 — strong for hard problems without a giant card.

16,384 context

Tech specs

deepseek-r1-distill-qwen-32b · 4,096 max output

Typical hosted: $0.22 / $0.65

$0.15 in / $0.45 out

per 1M tokens

~30% cheaper

Available on request

Expected 16 tok/s

DeepSeek R1 Distill 8B

Compact reasoning model that fits mid-range gaming GPUs.

16,384 context

Tech specs

deepseek-r1-distill-llama-8b · 4,096 max output

Typical hosted: $0.05 / $0.12

$0.03 in / $0.08 out

per 1M tokens

~35% cheaper

Available on request

Expected 30 tok/s

Mistral Small 24B

Mistral's efficient open instruct model for chat, drafting, and light agents.

32,768 context

Tech specs

mistral-small-24b-instruct · 4,096 max output

Typical hosted: $0.15 / $0.40

$0.10 in / $0.30 out

per 1M tokens

~30% cheaper

Available on request

Expected 22 tok/s

Gemma 3 12B IT

Google's open instruction-tuned model — solid chat and multilingual coverage.

8,192 context

Tech specs

gemma-3-12b-it · 2,048 max output

Typical hosted: $0.08 / $0.22

$0.05 in / $0.15 out

per 1M tokens

~35% cheaper

Available on request

Expected 28 tok/s

Gemma 3 27B IT

Larger Gemma 3 instruct variant for harder reasoning and longer answers.

8,192 context

Tech specs

gemma-3-27b-it · 2,048 max output

Typical hosted: $0.18 / $0.50

$0.12 in / $0.35 out

per 1M tokens

~30% cheaper

Available on request

Expected 18 tok/s

Phi-4

Microsoft's compact high-quality open model — punchy reasoning in a small footprint.

16,384 context

Tech specs

phi-4 · 2,048 max output

Typical hosted: $0.06 / $0.18

$0.04 in / $0.12 out

per 1M tokens

~35% cheaper

Available on request

Expected 32 tok/s

Mixtral 8x7B Instruct

Sparse MoE classic — strong open generalist that many tools already know.

32,768 context

Tech specs

mixtral-8x7b-instruct · 2,048 max output

Typical hosted: $0.18 / $0.50

$0.12 in / $0.35 out

per 1M tokens

~30% cheaper

Available on request

Expected 20 tok/s

Call it like OpenAI

Same request shape. Open models. No GPU ops on your side.

curl https://api.thirdshift.dev/v1/chat/completions \
  -H "Authorization: Bearer $THIRDSHIFT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen2.5-7b-instruct","messages":[{"role":"user","content":"Write one launch note."}],"stream":false}'