OpenAI GPT-6.1 Sol: $2/$10 pricing + 95% cached discount — released Oct 1, 2026
OpenAI GPT-6.1 Sol launched Oct 1, 2026: $2/$10 pricing per 1M tokens + 95% cached discount. Mid-tier between GPT-6 mini and Astra. AAII 52, 256K context.
OpenAI launched GPT-6.1 Sol on Oct 1, 2026 — a mid-tier production model priced at $2 input / $10 output per 1M tokens with a 95% cached-input discount ($0.10 cached input). Sol sits between GPT-6 mini ($0.30/$1.20) and GPT-6 Astra ($4/$16) in OpenAI's lineup, optimized for high-volume production workloads where prompt caching is dominant.
This is OpenAI's most aggressive pricing move in 2026 — a direct response to Anthropic Claude Sonnet 5.5 ($3/$15) and Google Gemini 4 Argon ($4/$20). With 95% cached discount, Sol becomes the cheapest production-tier model for cached workloads — underpricing every competitor by 50-90%.
What is OpenAI GPT-6.1 Sol?
GPT-6.1 Sol is a mid-tier production model in OpenAI's GPT-6 family, positioned between GPT-6 mini (cheaper, weaker) and GPT-6 Astra (expensive, strongest). It's a non-reasoning model — Sol is fast-tuned for direct answers, not chain-of-thought reasoning. For reasoning workloads, OpenAI recommends o4 or GPT-6 Astra.
Key positioning: Sol targets high-volume production workloads — chat assistants, content pipelines, search indexing, summarization, classification — where latency and cached prompt reuse are critical.
| Aspect | Value |
|---|---|
| Model ID | gpt-6.1-sol |
| Release date | Oct 1, 2026 |
| Positioning | Mid-tier production |
| Reasoning support | No (direct answers only) |
| Context window | 256K tokens |
| Knowledge cutoff | March 2026 |
| Latency | ~280ms TTFT (time to first token) |
| Throughput | ~500 tokens/sec sustained |
What does Sol cost?
Sol's pricing is the headline news:
| Item | Price per 1M tokens |
|---|---|
| Input | $2.00 |
| Cached input | $0.10 (95% discount) |
| Output | $10.00 |
Cached input pricing is the breakthrough: Sol can return cached prompt portions at $0.10 per 1M tokens — a 95% discount vs the standard input price. For workloads with large system prompts that don't change (chat assistants, RAG with stable contexts, classification with consistent instructions), cached pricing reduces effective input cost to $0.10 per 1M tokens.
Cost example: a customer-support assistant with a 10K-token system prompt + 5K-token user query + 200-token response, run 100,000 times:
- Standard input pricing: (10K + 5K) × $2 / 1M × 100K = $3,000 input
- Cached input pricing: 10K × $0.10 / 1M × 100K + 5K × $2 / 1M × 100K = $100 cached + $1,000 new = $1,100 input
- Savings: $1,900 (63%)
For workloads where system prompts are 50K+ tokens (heavy agents, complex RAG), savings reach 80-90%.
How does Sol compare to other models?
Sol vs other OpenAI models
| Model | Input | Output | Cached input | Intelligence Index |
|---|---|---|---|---|
| GPT-6 nano | $0.10 | $0.40 | $0.05 | 28 |
| GPT-6 mini | $0.30 | $1.20 | $0.15 | 38 |
| GPT-6.1 Sol | $2.00 | $10.00 | $0.10 | 52 |
| GPT-6 Astra | $4.00 | $16.00 | $0.40 | 53 |
| o4 (reasoning) | $5.00 | $20.00 | $0.50 | 64 |
Sol at 52 on Intelligence Index is 1 point below GPT-6 Astra (53) and 14 points above GPT-6 mini (38). Pricing undercuts Astra by 50% on input, 37% on output, and matches mini on cached input — making Sol the best intelligence-per-dollar in OpenAI's lineup.
Sol vs competitor models
| Model | Input | Output | Cached input | Intelligence Index |
|---|---|---|---|---|
| GPT-6.1 Sol | $2.00 | $10.00 | $0.10 | 52 |
| Claude Sonnet 5.5 | $3.00 | $15.00 | $0.30 | 54 |
| Gemini 4 Argon (intro) | $2.00 | $10.00 | — | 53 |
| Gemini 4 Argon (standard) | $4.00 | $20.00 | — | 53 |
| Grok 4.7 | $5.00 | $15.00 | — | 51 |
| Meta Muse Spark 1.4 | $1.50 | $8.00 | — | 47 |
Sol wins on cached input (95% discount is unmatched) and ties Gemini 4 Argon intro on input price. Sol loses to Sonnet 5.5 on intelligence (52 vs 54). For cached-dominant production workloads, Sol is the cheapest top-3 intelligence model available.
What are Sol's benchmarks?
Artificial Analysis Intelligence Index: 52 (vs GPT-6 Astra's 53, GPT-6 mini's 38).
Key benchmarks (per OpenAI's model card):
| Benchmark | Sol | GPT-6 Astra | GPT-6 mini |
|---|---|---|---|
| MMLU Pro | 86.1% | 92.4% | 73.2% |
| GPQA Diamond | 78.3% | 84.1% | 58.6% |
| HumanEval+ | 88.2% | 91.7% | 67.4% |
| MATH | 84.9% | 89.3% | 61.8% |
| LiveCodeBench v6 | 71.2% | 77.9% | 48.7% |
| AAII Intelligence Index | 52 | 53 | 38 |
Sol is within 5-10% of GPT-6 Astra on most benchmarks while costing 50% less on input and 37% less on output. For most production workloads, Sol is the best cost-quality trade-off in OpenAI's lineup.
What workloads is Sol best for?
Sol's sweet spot is high-volume production workloads with large reusable system prompts — exactly the workloads where cached input pricing matters most:
Best for:
- Customer support chatbots — large system prompt + variable user query + short response
- RAG applications — fixed retrieval context + variable question + short answer
- Content pipelines — fixed style guide + variable input + generated output
- Classification & tagging — fixed category list + variable content + label
- Search & ranking — fixed corpus summary + variable query + ranked results
- Code completion — fixed codebase context + variable code + completion
Not ideal for:
- Long reasoning chains — Sol is non-reasoning; use o4 or Astra for complex multi-step reasoning
- One-shot Q&A with small prompts — cached discount doesn't help; consider GPT-6 mini
- Frontier research — Sol is 1 point below Astra on AAII; for the absolute frontier, use Astra
How does Sol's cached discount compare?
Cached input discounts across major labs:
| Provider | Standard input | Cached input | Discount |
|---|---|---|---|
| OpenAI Sol | $2.00 | $0.10 | 95% |
| OpenAI GPT-6 mini | $0.30 | $0.15 | 50% |
| OpenAI GPT-6 Astra | $4.00 | $0.40 | 90% |
| Anthropic Claude Sonnet 5.5 | $3.00 | $0.30 | 90% |
| Anthropic Claude Haiku 4.5 | $0.80 | $0.08 | 90% |
| Google Gemini 4 Argon | $2.00 (intro) | Not disclosed | — |
| Google Gemini 4 Flash | $0.15 | $0.075 | 50% |
Sol's 95% cached discount is the deepest in the industry. Combined with Sol's already-low $2 base input price, cached Sol is the cheapest high-quality model per $1 in 2026.
Is Sol worth switching to?
Depends on workload:
- If you currently use GPT-6 Astra for high-volume production — Sol will save 50-80% on input cost with ~3-5% quality loss. Worth it for most production workloads.
- If you currently use GPT-6 mini — Sol will cost ~6x more but give 37% higher AAII. Worth it for production-critical workloads where quality matters.
- If you currently use Claude Sonnet 5.5 — Sol undercuts by 33% on input, 33% on output, with deeper cached discount. Worth switching for cost-sensitive workloads.
- If you currently use Gemini 4 Argon (intro) — Sol ties on input pricing, beats on cached discount. Sol is the better choice for cached-dominant workloads.
- If you need long reasoning — Sol is non-reasoning. Use o4 or GPT-6 Astra.
How does Sol fit in the October 2026 AI launch wave?
Sol is part of the October 2026 AI launch wave — alongside Aleph Alpha Kolibri (Oct 4), Anthropic Claude for Government GA (Oct 1), Gemini 4 Argon rollout (Sept 30–Oct 1), and 20+ other specialist models. For full context, see our October 2026 AI launches recap.
Sol is the only major frontier-class launch in October 2026 — everything else is specialist (voice, decision, sovereign). It signals OpenAI's strategy: dominate the cost-quality frontier while other vendors focus on niche verticals.
FAQ: OpenAI GPT-6.1 Sol
When was GPT-6.1 Sol released?
Oct 1, 2026. Available via OpenAI API immediately. ChatGPT availability: Pro / Team / Enterprise only at launch.
How much does GPT-6.1 Sol cost?
$2 input / $10 output per 1M tokens. Cached input: $0.10 per 1M tokens (95% discount).
Is Sol a reasoning model?
No — Sol is a direct-answer model. For reasoning, use OpenAI o4 or GPT-6 Astra.
What is the cached input discount?
When you re-use prompt portions that OpenAI has cached from a previous request (within 5-10 minutes), those tokens are billed at $0.10 per 1M instead of $2.00 — a 95% discount.
How does Sol compare to GPT-6 Astra?
52 vs 53 on AAII (within 5-10% on most evals) at half the input price and 63% of the output price. Sol is the better cost-quality trade-off for production workloads.
How does Sol compare to Claude Sonnet 5.5?
52 vs 54 on AAII — Sol is 2 points behind. Sol undercuts Sonnet 5.5 by 33% on input, 33% on output, and has a deeper cached discount (95% vs 90%).
What's Sol's context window?
256K tokens. Same as GPT-6 mini and Astra.
What's Sol's latency?
~280ms time to first token. ~500 tokens/sec sustained throughput.
Can I use Sol in ChatGPT?
Pro, Team, and Enterprise at launch. Free and ChatGPT Edu availability expected in Q4.
Does Sol support function calling?
Yes — full function calling, structured outputs, vision (image), and audio input support. No video input.
What's the difference between Sol and GPT-6 mini?
Sol is 14 AAII points higher (52 vs 38) at 6x the cost. Choose Sol when you need near-Astra quality without Astra pricing.
Why did OpenAI launch Sol?
Competitive response to Claude Sonnet 5.5 and Gemini 4 Argon. Sol undercuts both on cached input pricing while matching Argon's intro price on input.
Will Sol pricing change?
OpenAI has hinted that $2/$10 is an "introductory price" that may rise after 6-12 months, mirroring Gemini 4 Argon's transition from $2/$10 intro to $4/$20 standard.
Bottom line: GPT-6.1 Sol launched Oct 1, 2026 at $2/$10 per 1M tokens with a 95% cached input discount — the deepest cached discount in the industry. 52 AAII points (vs Astra's 53, Sonnet 5.5's 54), 256K context, 280ms TTFT. Best cost-quality trade-off for high-volume production workloads with large reusable system prompts.
Related articles
Picked from articles covering the same devices and topics.
Minecraft Sift mod: bring LLMs into Minecraft via MCP — install + Claude/OpenAI guide (2026)
Minecraft Sift mod brings LLMs into Minecraft via Model Context Protocol (MCP). Install guide, supported models (Claude, GPT, local), agent examples. 2026 viral mod.
Meta Muse AI: 3 models in 3 weeks — Spark 1.3 (Sept 2), Glimmer (Aug), Muse agent (Sept 8)
Meta Muse AI: 3 products in Aug-Sept 2026 — Spark 1.3 (Sept 2, beats GPT-5.6 Sol on coding), Glimmer (Aug, open-weight), and Muse agent (Sept 8, FB/IG/WhatsApp). Watermelon in dev.
iPhone Duo specs: 5.4" cover + 7.6" inner, 44h/31h battery, $1,999 (Oct 23)
iPhone Duo specs (Apple's first foldable, $1,999, ships Oct 23, 2026): 5.4" cover + 7.6" inner display, 44h outer / 31h inner video, A20 Pro, 2 colors (Night Sky + Star White). Touch ID, IP68.
Sources
(4 references)- [1]OpenAI — GPT-6.1 Sol launch announcement(2026-10-01)↩
- [2]OpenAI Pricing — GPT-6.1 Sol rates(2026-10-01)↩
- [3]Simon Willison — GPT-6.1 Sol review(2026-10-02)↩
- [4]Artificial Analysis — GPT-6.1 Sol benchmarks(2026-10-02)↩