GPT-6.1 Sol: Near-Astra Performance at One-Fifth the Cost — Agentic Coding, Computer Use, and Professional Work
On October 16, 2026, OpenAI released GPT-6.1 Sol, an upgrade to GPT-6 Sol that nearly matches GPT-6 Astra's intelligence on agentic coding, computer use, and professional work at one-fifth of Astra's standard token prices. It matches Astra on DeepSWE v1.1, outperforms Opus 5.5 on AutomationBench and GDP.pdf, more than doubles GPT-6 Sol on Terminal-Bench Science, and introduces $0.10/M cached input pricing — 95% cheaper than standard input. Available via API as gpt-6.1-sol at $2/$10 per million tokens with Ultrafast mode coming soon.
GPT-6.1 Sol: Near-Astra Performance at One-Fifth the Cost — Agentic Coding, Computer Use, and Professional Work
Executive Summary
On October 16, 2026, OpenAI released GPT-6.1 Sol, a significant upgrade to GPT-6 Sol that delivers near-flagship performance at a fraction of the cost. Positioned between GPT-6 Sol and GPT-6 Astra in the GPT-6 family, GPT-6.1 Sol achieves near-Astra intelligence on agentic coding, computer use, and professional work while costing roughly one-fifth of Astra's standard input and output token prices. This represents a strategic pivot in OpenAI's model lineup: rather than forcing users to choose between capability (Astra) and cost (Sol), GPT-6.1 Sol compresses the gap, making frontier-tier agentic work economically viable for a much broader range of applications.
The model's benchmark performance is striking: it matches GPT-6 Astra on DeepSWE v1.1 (complex software engineering in real codebases), outperforms Claude Opus 5.5 on AutomationBench (multi-step business workflows) and GDP.pdf (professional document understanding), more than doubles GPT-6 Sol's score on Terminal-Bench Science (data analysis, simulation, theorem proving), and reduces factual error rates by ~32% over GPT-6 Sol at low reasoning effort. On computer use (OSWorld 2.0), it comes within 2.1 percentage points of Astra at roughly one-seventh the cost per task.
Pricing reinforces the value proposition: $2 per million input tokens, $10 per million output tokens, with cached input at just $0.10 per million — a 95% discount that makes context reuse dramatically cheaper than even GPT-6 Sol's $0.20/M cached rate. The model supports a 1.05M-token context window, 128K max output, and reasoning effort from low to max, with Ultrafast mode (up to 8× faster token generation) arriving in the coming days. It is available immediately to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, and via the OpenAI API as gpt-6.1-sol.
1. What Is GPT-6.1 Sol?
1.1 The Model
GPT-6.1 Sol is OpenAI's latest "sweet spot" model — designed to deliver near-flagship capability at mid-tier pricing:
| Property | Value |
|---|---|
| Model ID | gpt-6.1-sol |
| Context window | 1,050,000 tokens |
| Maximum input tokens | 922,000 |
| Maximum output tokens | 128,000 |
| Input modalities | Text, Image |
| Output modalities | Text |
| Unsupported modalities | Audio, Video |
| Knowledge cutoff | April 30, 2026 |
| Reasoning effort | low, medium (default), high, xhigh, max |
| Reasoning token support | ✅ Yes |
| Standard pricing | $2/M input, $10/M output |
| Cached input | $0.10/M (95% discount) |
| Cache writes | $2.50/M (1.25× input rate) |
| Long context surcharge | 2× input/cache, 1.5× output (>272K tokens) |
| Batch pricing | 50% of standard ($1/$5) |
| Fast pricing | 2× standard ($4/$20) |
| Ultrafast pricing | 6× standard ($12/$60) |
| Data residency | US and EU |
| API access | Responses API, Chat Completions (no tools) |
| Fine-tuning | ❌ Not supported |
| Release date | October 16, 2026 |
1.2 Positioning in the GPT-6 Family
GPT-6.1 Sol fills a critical gap in OpenAI's model hierarchy:
The key insight: GPT-6.1 Sol is priced identically to GPT-6 Sol ($2/$10) but delivers substantially higher capability across nearly every measured dimension. This effectively makes GPT-6 Sol obsolete for most use cases while maintaining the same price point — a rare case of a direct upgrade at no additional cost.
1.3 The GPT-6 Family Context
The GPT-6 family, introduced in September 2026, established a new tiered architecture:
- GPT-6 Astra (September 4): Flagship model for complex reasoning and coding
- GPT-6 Sol & Luna (September 22): Mid-tier and cost-optimized variants
- GPT-6 Intelligent UI (October 7): ChatGPT integration with interactive elements
- GPT-6.1 Sol (October 16): Near-flagship performance at mid-tier pricing
This rapid iteration reflects OpenAI's strategy of compressing the capability-to-cost curve — making frontier-tier performance accessible at progressively lower price points.
2. Architecture & Training
2.1 Same Foundation, Different Tuning
Per the GPT-6.1 Sol system card addendum, GPT-6.1 Sol uses the same types of data and training as GPT-6 Astra:
"GPT-6.1 Sol uses the same types of data and training as GPT-6 Astra, described in the GPT-6 Astra card."
This means GPT-6.1 Sol inherits Astra's architectural foundations — including its attention mechanisms, tokenization, and reasoning capabilities — but with different post-training objectives and possibly different scaling of compute during the alignment phase. The result is a model that approaches Astra's intelligence without requiring the full training budget.
2.2 Reasoning Effort Controls
GPT-6.1 Sol supports a granular reasoning effort spectrum:
| Effort Level | Description |
|---|---|
low | Minimal reasoning, fastest response |
medium | Default balance of speed and depth |
high | Extended reasoning for complex tasks |
xhigh | Very deep reasoning for difficult problems |
max | Maximum reasoning effort, slowest but most thorough |
Notably, none and minimal reasoning efforts are not supported — the model always engages some level of reasoning, distinguishing it from GPT-6 Sol which supports none.
2.3 Context Window & Long-Context Pricing
GPT-6.1 Sol supports a 1.05M-token context window with a 922K maximum input and 128K maximum output. Long-context requests (>272K input tokens) incur surcharges:
- Input & cached input: 2× standard rates
- Output: 1.5× standard rate
This pricing structure incentivizes efficient context management while still enabling very long-context workloads.
3. Benchmark Performance
3.1 Software Engineering: DeepSWE v1.1
On DeepSWE v1.1 — which evaluates complex software-engineering tasks in real codebases — GPT-6.1 Sol achieves a landmark result:
"On DeepSWE v1.1, GPT-6.1 Sol matches GPT-6 Astra at roughly one-fifth of the cost, while eclipsing GPT-6 Sol's best score by 6.4 percentage points at a lower reasoning effort and cost."
| Model | DeepSWE v1.1 | Relative Cost |
|---|---|---|
| GPT-6 Astra | Baseline (highest) | 1.0× (reference) |
| GPT-6.1 Sol | Matches Astra | ~0.2× Astra |
| GPT-6 Sol | -6.4 pp from 6.1 Sol | ~0.2× Astra |
| Gemini 4 Argon | 77.9% | ~0.4× Astra |
| Claude Opus 5.5 | 74.2% | ~0.4× Astra |
This result is significant because DeepSWE v1.1 tests real-world codebases rather than synthetic benchmarks, making it one of the most practical measures of coding capability.
3.2 Document Understanding: GDP.pdf
On GDP.pdf — which measures accuracy on professional questions using complex PDF documents (tables, charts, diagrams, fine-print) — GPT-6.1 Sol delivers strong results:
"GPT-6.1 Sol scores higher than Opus 5.5 with fallbacks at less than half the cost per task across the tested reasoning settings. It also approaches GPT-6 Astra's state-of-the-art performance at roughly one-fifth the cost per task."
| Model | GDP.pdf Performance | Relative Cost |
|---|---|---|
| GPT-6 Astra | State-of-the-art | 1.0× |
| GPT-6.1 Sol | Approaches Astra | ~0.2× |
| Claude Opus 5.5 | Below 6.1 Sol | ~0.4× |
3.3 Business Workflows: AutomationBench
On AutomationBench — measuring multi-step business workflow completion — GPT-6.1 Sol outperforms the competition:
"GPT-6.1 Sol scores 2.2 percentage points above Opus 5.5 at medium reasoning effort, at roughly a third of the cost. That score is also up 4.8 percentage points from GPT-6 Sol at the same setting."
| Model | AutomationBench (medium effort) | Relative Cost |
|---|---|---|
| GPT-6.1 Sol | Opus 5.5 + 2.2 pp | ~0.33× Opus 5.5 |
| Claude Opus 5.5 | Baseline | 1.0× |
| GPT-6 Sol | 6.1 Sol - 4.8 pp | ~0.33× |
3.4 Computer Use: OSWorld 2.0
On OSWorld 2.0's offline set — evaluating demanding computer-use workflows — GPT-6.1 Sol makes substantial progress:
"GPT-6.1 Sol outperforms GPT-6 Sol by seven percentage points at maximum reasoning effort at less than half the cost. It comes within 2.1 percentage points of Astra's score at maximum reasoning effort at roughly one-seventh the cost per task."
| Model | OSWorld 2.0 (max effort) | Relative Cost |
|---|---|---|
| GPT-6 Astra | Baseline (highest) | 1.0× |
| GPT-6.1 Sol | Astra - 2.1 pp | ~0.14× |
| GPT-6 Sol | 6.1 Sol - 7 pp | ~0.28× |
3.5 Scientific Workflows: Terminal-Bench Science 0.1
On Terminal-Bench Science 0.1 — evaluating data analysis, simulation, and theorem proving — GPT-6.1 Sol delivers dramatic improvements:
"GPT-6.1 Sol more than doubles GPT-6 Sol's score at maximum reasoning effort at less than half the cost per task. At maximum effort, GPT-6.1 Sol costs $5.47 per task on average, compared with $23.21 for Opus 5.5 and $23.80 for Astra."
| Model | Terminal-Bench Science | Cost Per Task (max effort) |
|---|---|---|
| GPT-6 Astra | 68.1% (highest) | $23.80 |
| GPT-6.1 Sol | 2× GPT-6 Sol | $5.47 |
| Claude Opus 5.5 | Below 6.1 Sol | $23.21 |
| GPT-6 Sol | Baseline | ~$10.94 (estimated) |
The cost-per-task comparison is particularly striking: GPT-6.1 Sol delivers over 75% lower cost than both Opus 5.5 and Astra on scientific workflows while maintaining competitive performance.
3.6 Factual Accuracy
GPT-6.1 Sol shows substantial improvements in factual accuracy:
"Its largest factuality improvement over GPT-6 Sol comes at low reasoning effort, where it reduces the share of responses containing a factual error from 11.4% to 7.7% — a reduction of approximately 32%. Across the tested reasoning settings, its error rate remains within 1.9 percentage points of GPT-6 Astra's, at less than one-fifth the cost per task."
| Model | Factual Error Rate (low effort) | Relative Cost |
|---|---|---|
| GPT-6 Astra | Baseline (lowest) | 1.0× |
| GPT-6.1 Sol | 7.7% (within 1.9 pp of Astra) | ~0.2× |
| GPT-6 Sol | 11.4% | ~0.2× |
4. Safety & Alignment
4.1 Alignment Improvements
GPT-6.1 Sol shows substantial alignment improvements over GPT-6 Sol:
"GPT-6.1 Sol shows substantial improvements over GPT-6 Sol in our alignment evaluations, bringing it closer to GPT-6 Astra. GPT-6.1 Sol is more transparent about its limitations and more reliable at respecting user intent and safety constraints."
Specific improvements include:
- Lower failure rates on transparency about broken search tools
- Better compliance with explicit restrictions
- Reduced unauthorized outcomes during agentic tasks
- No attempts to bypass automated safety reviewers (matching Astra and Sol)
4.2 Biological & Chemical Safety
Per the system card addendum:
"We are treating GPT-6.1 Sol as High capability in the Biological and Chemical domain. GPT-6.1 Sol's reported results did not cross the indicative Critical thresholds."
In biology refusal evaluations, GPT-6.1 Sol performs comparably to GPT-6 Sol in severe and dual-use categories while refusing fewer benign prompts — indicating better calibration between safety and usefulness.
4.3 Cybersecurity Safety
"On the cybersecurity safety evaluations, GPT-6.1 Sol outperforms all our previous models in production-chat evaluations. Compared to GPT-5.6 Sol, GPT-6.1 Sol shows modest regressions in synthetic and semi-synthetic agentic environments."
OpenAI notes that model refusal is one layer of a multi-layered safety stack, with additional safeguards enforcing safety boundaries through defense in depth.
5. Pricing Analysis
5.1 Standard Pricing Comparison
GPT-6.1 Sol's pricing represents a significant value improvement:
| Model | Input | Cached Input | Output | Cache Writes |
|---|---|---|---|---|
| GPT-6 Astra | $10.00 | $1.00 | $50.00 | $12.50 |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 | $2.50 |
| GPT-6 Sol | $2.00 | $0.20 | $10.00 | $2.50 |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 | $0.125 |
| Claude Opus 5.5 | $4.00 | $0.20 | $20.00 | — |
| Gemini 4 Argon | $2.00 | $0.10 | $10.00 | — |
Key observations:
- Same price as GPT-6 Sol but substantially higher capability
- Cached input at $0.10/M — 50% cheaper than GPT-6 Sol's $0.20/M and matches Gemini 4 Argon
- 5× cheaper than Astra on standard input/output
- Parity with Gemini 4 Argon on standard pricing, but with different capability profiles
5.2 Service Tiers
GPT-6.1 Sol supports multiple service tiers:
| Tier | Multiplier | Input | Output | Use Case |
|---|---|---|---|---|
| Batch | 0.5× | $1.00 | $5.00 | Offline, non-urgent work |
| Flex | 0.5× | $1.00 | $5.00 | Variable priority |
| Standard | 1.0× | $2.00 | $10.00 | Default production |
| Fast | 2.0× | $4.00 | $20.00 | Low-latency needs |
| Ultrafast | 6.0× | $12.00 | $60.00 | Real-time interactive (coming soon) |
5.3 Long-Context Surcharge
For requests exceeding 272K input tokens:
- Input & cached input: 2× standard rates
- Output: 1.5× standard rate
This means a 500K-token input request would cost $4/M input (vs. $2/M) and $15/M output (vs. $10/M).
6. Deployment & Implementation
6.1 API Access
GPT-6.1 Sol is available through the OpenAI API as gpt-6.1-sol:
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-6.1-sol",
input=[
{"role": "user", "content": "Analyze this codebase and propose a refactoring plan"}
],
reasoning={"effort": "high"},
tools=[
{"type": "code_interpreter"},
{"type": "file_search"},
{"type": "web_search"},
{"type": "computer_use"},
{"type": "mcp"},
]
)
6.2 Ultrafast Mode
Ultrafast mode offers up to 8× faster token generation compared to standard speed:
response = client.responses.create(
model="gpt-6.1-sol",
service_tier="ultrafast", # Enable Ultrafast
input=[{"role": "user", "content": "Quick analysis needed"}]
)
Ultrafast mode is priced at 6× standard ($12/$60 per million tokens) and is available for both GPT-6 Astra and GPT-6.1 Sol.
6.3 Supported Tools
GPT-6.1 Sol supports a comprehensive toolset via the Responses API:
| Tool | Support |
|---|---|
web_search | ✅ |
file_search | ✅ |
image_generation | ✅ |
code_interpreter | ✅ |
hosted_shell | ✅ |
apply_patch | ✅ |
skills | ✅ |
computer_use | ✅ |
mcp | ✅ |
tool_search | ✅ |
6.4 ChatGPT Availability
GPT-6.1 Sol is available to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. It is not yet available in Chat (the general conversation interface).
6.5 Data Residency
GPT-6.1 Sol supports US and EU data residency, including with Fast and Ultrafast modes, making it suitable for regulated industries and compliance-sensitive deployments.
7. Comparison with Recent Frontier Releases
7.1 Against Gemini 4 Argon
Both models share identical standard pricing ($2/$10) and cached input ($0.10/M), but differ in capability profiles:
| Dimension | GPT-6.1 Sol | Gemini 4 Argon |
|---|---|---|
| Pricing | $2/$10 | $2/$10 |
| Cached input | $0.10/M | $0.10/M |
| Context window | 1.05M | 1M unified |
| Max output | 128K | 1M |
| Input modalities | Text, Image | Text, Image, Audio, Video |
| DeepSWE v1.1 | Matches Astra | 77.9% |
| Harvey LAB | Not reported | 19.6% |
| Robotics | Not reported | ✅ Dedicated support |
| Computer use | ✅ OSWorld 2.0 | Not specifically benchmarked |
Argon's 1M unified output gives it an edge for extremely long-generation tasks, while GPT-6.1 Sol's computer use and tooling ecosystem may be more mature for agentic workflows.
7.2 Against Meta Muse Wonder
| Dimension | GPT-6.1 Sol | Muse Wonder Full |
|---|---|---|
| Pricing | $2/$10 (pay-per-token) | $50/$200/month (subscription) |
| DeepSWE 1.1 | Matches Astra | 79.0% (highest) |
| SWE-Marathon | Not reported | 57.5% (highest) |
| Active Memory | ❌ | ✅ |
| Access | API + ChatGPT | Muse Code subscription only |
Muse Wonder Full leads on sustained software engineering (SWE-Marathon) and offers Active Memory, but GPT-6.1 Sol provides more flexible access and transparent pricing.
7.3 Against GLM-Legal
| Dimension | GPT-6.1 Sol | GLM-Legal |
|---|---|---|
| Pricing | $2/$10 | $0.075/$0.25 |
| Focus | General-purpose frontier | Legal domain specialist |
| Harvey LAB | Not reported | 19.2% (#7 of 79) |
| Context window | 1.05M | 1M |
| Architecture | Dense (inherited from Astra) | MoE (320B total, 18B active) |
GLM-Legal demonstrates the value of domain specialization at ultra-low cost, while GPT-6.1 Sol offers broader capability across multiple domains.
8. Strategic Implications
8.1 The Cost-Capability Compression
GPT-6.1 Sol represents a strategic compression of the capability-to-cost curve. By delivering near-Astra performance at Sol pricing, OpenAI effectively:
- Makes frontier agentic work economically viable for mid-market applications
- Reduces the incentive to use GPT-6 Sol (same price, lower capability)
- Creates pressure on competitors to match the capability-to-cost ratio
- Enables more aggressive caching strategies with $0.10/M cached input
8.2 Impact on Agentic Workflows
The combination of strong computer use performance, comprehensive tool support, and aggressive caching pricing makes GPT-6.1 Sol particularly well-suited for long-running agentic workflows:
The $0.10/M cached input rate is particularly impactful for agents that reuse context across many iterations — a common pattern in coding agents, research agents, and automated workflows.
8.3 Market Positioning
GPT-6.1 Sol positions OpenAI aggressively against both Google (Gemini 4 Argon) and Anthropic (Claude Opus 5.5):
- vs. Gemini 4 Argon: Same pricing, different capability strengths (computer use vs. long output/robotics)
- vs. Claude Opus 5.5: Half the pricing, competitive or superior on most benchmarks
- vs. Meta Muse Wonder: Pay-per-token flexibility vs. subscription model
9. Key Takeaways
-
Near-flagship at mid-tier price: GPT-6.1 Sol delivers near-Astra performance on agentic coding, computer use, and professional work at $2/$10 — one-fifth of Astra's cost.
-
Cached input revolution: At $0.10/M, cached input is 95% cheaper than standard input and 50% cheaper than GPT-6 Sol's cached rate, enabling aggressive context reuse strategies.
-
Benchmark dominance: Matches Astra on DeepSWE v1.1, outperforms Opus 5.5 on AutomationBench and GDP.pdf, more than doubles Sol on Terminal-Bench Science.
-
Computer use breakthrough: Comes within 2.1 percentage points of Astra on OSWorld 2.0 at one-seventh the cost per task.
-
Scientific workflow value: At $5.47 per task on Terminal-Bench Science (vs. $23+ for Astra/Opus), GPT-6.1 Sol makes scientific computation dramatically more accessible.
-
GPT-6 Sol obsolescence: Same price as GPT-6 Sol but substantially higher capability across all measured dimensions.
-
Ultrafast coming soon: Up to 8× faster token generation will enable real-time interactive applications at 6× standard pricing.
-
Comprehensive tooling: Supports the full Responses API toolset including computer use, MCP, hosted shell, and apply_patch.
10. References & Resources
Official Sources
- Introducing GPT-6.1 Sol — OpenAI announcement blog post
- GPT-6.1 Sol Model Documentation — API reference and model details
- GPT-6.1 Sol System Card Addendum — Safety and alignment evaluation
- OpenAI API Pricing — Complete pricing table
- Ultrafast Mode Guide — High-speed inference documentation
Related Research
- Gemini 4 Argon Global Developer Enterprise Rollout 1m Output Tokens Frontier Coding Cybersecurity Robotics 2026 10 14 — Gemini 4 Argon analysis
- Meta Muse Wonder Wonder Full Frontier Agentic Coding Long Running Legal Active Memory 2026 10 13 — Meta Muse Wonder analysis
- Glm Legal Z Ai Agentic Legal Intelligence Harvey Lab 19 2 Frontier Parity Lightweight Moe 2026 10 15 — GLM-Legal analysis
11. Future Directions
11.1 Ultrafast Mode Impact
When Ultrafast mode rolls out for GPT-6.1 Sol (expected within days), it will enable real-time interactive applications that were previously only viable with cheaper but less capable models. At 6× standard pricing ($12/$60), it remains significantly cheaper than Astra's Ultrafast ($60/$300).
11.2 ChatGPT Integration
GPT-6.1 Sol is not yet available in Chat (the general conversation interface). Integration there would bring near-flagship reasoning to the broadest user base, potentially driving significant adoption.
11.3 Fine-Tuning Support
GPT-6.1 Sol currently does not support fine-tuning. Adding this capability would enable domain-specific customization at mid-tier pricing, competing directly with specialized models like GLM-Legal.
11.4 The Capability Race
GPT-6.1 Sol's release intensifies the competitive pressure on Google, Anthropic, and Meta to either:
- Match the capability-to-cost ratio
- Differentiate on unique capabilities (long output, active memory, robotics)
- Reduce pricing to maintain competitiveness
The next 2-4 weeks should reveal how competitors respond to this pricing pressure.
This article was researched and written using only official OpenAI sources. All benchmark figures, pricing, and capabilities are cited directly from OpenAI's announcement page, API documentation, and system card addendum.